Astronomy
&
Astrophysics
A&A 588, A137 (2016)
DOI: 10.1051/0004-6361/201527821
c ESO 2016
Young and middle age pulsar light-curve morphology: Comparison
of Fermi observations with γ-ray and radio emission geometries
M. Pierbattista1,2 , A. K. Harding3 , P. L. Gonthier4 , and I. A. Grenier5,6
1
2
3
4
5
6
Nicolaus Copernicus Astronomical Center, Rabiańska 8, 87-100 Toruń, Poland
e-mail: [email protected]
INAF−Istituto di Astrofisica Spaziale e Fisica Cosmica, 20133 Milano, Italy
Astrophysics Science Division, NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA
Hope College, Department of Physics, Holland MI 49423, USA
Laboratoire AIM, Université Paris Diderot/CEA-IRFU/CNRS, Service d’Astrophysique, CEA Saclay, 91191 Gif-sur-Yvette, France
Institut Universitaire de France, 75231 Paris Cedex 05, France
Received 24 November 2015 / Accepted 18 January 2016
ABSTRACT
Thanks to the huge amount of γ-ray pulsar photons collected by the Fermi Large Area Telescope since its launch in June 2008, it is
now possible to constrain γ-ray geometrical models by comparing simulated and observed light-curve morphological characteristics.
We assumed vacuum-retarded dipole (VRD) pulsar magnetic field and tested simulated and observed morphological light-curve characteristics in the framework of two pole emission geometries, Polar Cap (PC) and Slot Gap (SG), and one pole emission geometries,
traditional Outer Gap (OG) and One Pole Caustic (OPC). Radio core plus cone emission was assumed for the pulsars of the simulated
sample. We compared simulated and observed recurrence of class shapes and peak multiplicity, peak separation, radio-lag distributions, and trends of peak separation and radio lag as a function of observable and non-observable pulsar parameters. We studied how
pulsar morphological characteristics change in multi-dimensional observable and non-observable pulsar parameter space.
The PC model gives the poorest description of the LAT pulsar light-curve morphology. The OPC best explains both the observed
γ-ray peak multiplicity and shape classes. The OPC and SG models describe the observed γ-ray peak-separation distribution for lowand high-peak separations, respectively. This suggests that the OPC geometry best explains the single-peak structure but does not
manage to describe the widely separated peaks predicted in the framework of the SG model as the emission from the two magnetic
hemispheres. The OPC radio-lag distribution shows higher agreement with observations suggesting that assuming polar radio emission, the γ-ray emission regions are likely to be located in the outer magnetosphere. Alternatively, the radio emission altitude could
be higher that we assumed. We compared simulated non-observable parameters with the same parameters estimated for LAT pulsars
in the framework of the same models. The larger agreement between simulated and LAT estimations in the framework of the OPC
suggests that the OPC model best predicts the observed variety of profile shapes. The larger agreement obtained between observations
and the OPC model predictions jointly with the need to explain the abundant 0.5 separated peaks with two-pole emission geometries,
calls for thin OPC gaps to explain the single-peak geometry but highlights the need of two-pole caustic emission geometry to explain
widely separated peaks.
Key words. stars: neutron – pulsars: general – gamma rays: stars – radiation mechanisms: non-thermal – methods: data analysis –
methods: statistical
1. Introduction
The successful launch of the Fermi Gamma-ray space telescope
satellite (hereafter Fermi) in June 2008 represents a milestone
in pulsar astrophysics research. Owing to the observations performed with its main instrument, the Large Area Telescope
(LAT), Fermi increased the number of known young or middleaged γ-rays pulsars by a factor ∼30, discovered the new category
of γ-ray millisecond pulsars, and obtained unprecedented high
quality light curves for the major part of the observed objects.
The increasing number of γ-ray pulsars detected by Fermi,
which is now second just to the number of pulsars detected
at radio wavelengths, offered the possibility, for the first time
with high statistics, of studying the collective properties of the
γ-ray pulsar population and to compare them with model predictions (e.g. pulsar population syntheses; Watters & Romani 2011;
Takata et al. 2011; light-curve fitting and morphology analysis; Watters et al. 2009; Romani & Watters 2010). Pierbattista
(2010) performed a comprehensive study of the Fermi γ-ray
pulsars detected by LAT during the first year of observations
(The first Fermi LAT catalogue of γ-ray pulsars, Abdo et al.
2010, hereafter PSRCAT1). Pierbattista (2010) tested radiative and geometrical models against the observations according to three different approaches: a neutron star (NS) population
synthesis to compare collective radiative properties of observed
and simulated pulsar populations; a simulation of the observed
pulsar emission patterns to estimate non-observable pulsar parameters and to study their relationship with observable pulsar
characteristics in the framework of different emission geometries; and a comparison of simulated and observed light-curve
morphological characteristics in the framework of different
emission geometries. The population synthesis and the estimate of non-observable LAT pulsar parameters have been presented in Pierbattista et al. (2012; hereafter PIERBA12) and by
Pierbattista et al. (2015; hereafter PIERBA15), respectively.
PIERBA12 synthesised a NS population, evolved this population in the galactic gravitational potential assuming a supernova kick velocity and birth space distribution, and computed,
Article published by EDP Sciences
A137, page 1 of 26
A&A 588, A137 (2016)
for each NS of the sample, γ-ray and radio radiation powers
and light curves according to radiative γ-ray models and a radio
model. The implemented radiative γ-ray models are: the Polar
Cap model (PC; Muslimov & Harding 2003), Slot Gap model
(SG; Muslimov & Harding 2004), Outer Gap model (OG; Cheng
et al. 2000), and the One Pole Caustic model (OPC; Romani &
Watters 2010; Watters et al. 2009), which is an alternative formulation of the OG that only differs in the computation of radiative power and width of the emission region. In the OPC model,
both gap width and gap-width luminosity have been set to reproduce the relation Lγ ∝ Ė 0.5 observed in PSRCAT1; since the gap
width is assumed equal to the γ-ray efficiency, wOPC ∝ Ė −0.5 ,
the luminosity must follow Lγ = ĖwOPC . Because of that, the
OPC cannot be considered a full physical model like PC, SG,
and OG, but a phenomenological formulation of the OG, built
in to reproduce the observations. Each pulsar radio luminosity
has been computed according to a radio core plus cone model,
following the prescriptions from Gonthier et al. (2004), Story
et al. (2007), and Harding et al. (2007). Simulation details can be
found in PIERBA12. The γ-ray and radio light curves of the simulated pulsars have been obtained according to the geometrical
model from Dyks et al. (2004). Under the assumption that γ-ray
and radio photons are emitted tangent to the poloidal magnetic
field line in the reference frame instantaneously corotating with
the pulsar, γ-ray and radio light curves have been computed by
assuming location and size of the emission region in the framework of each γ-ray model and radio model. The directions of the
photons generated at the emission point have been computed as
described in Bai & Spitkovsky (2010). The complete description
of the method used to simulate γ-ray and radio light curves of the
observed pulsars can be found in PIERBA15.
Pierbattista et al. (2015) simulated the emission pattern of
the young or middle-aged LAT γ-ray pulsars detected after three
years of observations and published in the second LAT γ-ray pulsars catalogue (Abdo et al. 2013, hereafter PSRCAT2), and used
them to fit the observed profiles. The LAT pulsar emission patterns were computed in the framework of the very same radiative
models implemented by PIERBA12, namely PC, SG, OG, OPC,
and radio core plus cone, and by assuming emission geometry
according to Dyks et al. (2004). For each LAT pulsar and each
model, γ-ray and radio light curves were computed for a grid of
values of magnetic obliquity α (angle between the pulsar magnetic and rotational axes) and observer line-of-sight ζ (angle between the rotational axis and the direction of the observer), both
stepped every 5◦ in the interval 1 to 90, for the actual pulsar spin
period (P), and for the magnetic field (B) and width of the emission gap region (W) computed as described in PIERBA12. A
best-fit light curve with relative best-fit parameters, α, ζ, and W,
were found in the framework of each model and for each pulsar
by fitting the observed light curves with the emission patterns
through a maximum likelihood criterion.
The aim of this paper is to further develop and improve the
light-curve morphological analysis implemented by Pierbattista
(2010) by testing the PC, SG, OG, OPC, and radio core plus
cone emission geometries against the young and middle-aged
pulsar sample published in PSRCAT2. We computed a series
of light-curve morphological characteristics, namely γ-ray and
radio peak numbers, their phase separation, the lag between
the occurrence of γ-ray and radio peaks in radio loud pulsars,
among others, for both observed and simulated light curves and
within each model. The simulated pulsar light curves are those
synthesised by PIERBA12 in the framework of PC, SG, OG,
OPC, and radio core plus cone models by assuming emission
geometry from Dyks et al. (2004). We build a light-curve shape
A137, page 2 of 26
classifications according to the recurrence of the morphological
characteristics in observed and simulated profiles and compared
simulated and observed shape classifications in the framework
of each model. We compared observed and simulated distributions of morphological characteristics as a function of observable and non-observable pulsar parameters. The non-observable
pulsar parameters, namely α, ζ, and γ-ray beaming factor fΩ ,
are those estimated by PIERBA15 in the framework of the implemented emission geometries. We obtained constraints on the
emission geometry that best explains the observations.
The outline of this paper is as follows. In Sect. 2 we characterise both simulated and observed γ-ray pulsar samples. In
Sect. 3 we describe the method used to classify simulated and
observed pulsars and compare simulated and observed γ-ray and
the peak multiplicity of radio light curves. In Sect. 4 we describe the criteria used to measure the light-curve morphological
characteristics for both observed and simulated objects and compare observed and simulated distributions and trends of observable and non-observable/estimated pulsar parameters. Section 5
shows the γ-ray peak multiplicity and peak separation in the
(α, ζ) plane for both simulated and observed pulsars (estimated
values). Section 6 summarises the results.
In Appendix A templates of the γ-ray and radio shape classes
defined in Sect. 3 are shown in the framework of each model.
In Appendix B the pulsar γ-ray and radio emission patterns are
shown to ease the interpretation of the results. In Appendix C
we compare the recurrence of simulated and observed γ-ray
light-curve multiplicity for radio quiet and radio loud pulsars
and of simulated and observed γ-ray and radio shape classes.
In Appendix D we give exhaustive maps showing the γ-ray peak
multiplicity and γ-ray peak separation as a function of the spin
period (PC) and width of the acceleration gap (SG, OG, and
OPC) in Figs. D.1 and D.2, respectively. Appendix E.1 describe
the one- and two-dimensional Kolmogorov-Smirnov (KS) tests
used to quantify the agreement between observed and simulated
distributions.
2. Data selection and simulated sample
We have analysed 77 young or middle-aged pulsars listed in
PSRCAT2. We obtained their γ-ray and radio light curves from
the LAT catalogue data products public link1 . We built each
Fermi γ-ray pulsar light curve with 3 years of LAT observations
from 2008 August 4 to 2011 August 4 selecting only photons
with energy larger than 100 MeV and belonging to the source
class, as defined in the P7_V6 instrument response function. We
obtained the Fermi γ-ray light curves as weighted light curves
where each photon is characterised by its probability to belong to
the pulsar with the photon weights following the Gaussian probability distribution (PSRCAT2). We built the radio light curves
of the analysed RL pulsars from observations mainly performed
at 1400 MHz by the radio telescopes operating within the Fermi
Pulsar Search Consortium (PSC; Ray et al. 2012) and the Fermi
Pulsar Timing Consortium (PTC; Smith et al. 2008), namely the
Green Bank Telescope (GBT), Parkes Telescope, Nançay Radio
Telescope (NRT), Arecibo Telescope, the Lovell Telescope at
Jodrell Bank, and the Westerbork Synthesis Radio Telescope
(Smith et al. 2008). See PSRCAT2 for details about how the
γ-ray and radio light curves were obtained.
The simulated radio and γ-ray light curves used in this paper are those obtained within the population synthesis from
1
http://fermi.gsfc.nasa.gov/ssc/data/access/lat/2nd_
PSR_catalog/
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
Table 1. Values adopted in each model for the light-curve smoothing σν and δσν , the zero threshold T h0 and maximum-minimum detection
threshold T hmx−mn , and the peak widths, WH and WB for modelled γ-ray and radio light curves.
σν
δσν
T h0
T hmx−mn
WH
WB
PC
SG
OG
OPC
Radio
0.005
0.001
0.02
0.001
W(0.50(max − min))
W(0)
0.9
0.1
0.05
0.15
W(min +0.50(max − min))
W(min +0.15(max − min))
0.2
0.01
0.03
0.03
W(0.50(max − min))
W(0)
0.2
0.01
0.03
0.03
W(0.50(max − min))
W(0)
0.2
0.01
0.04
0.01
W(0.50(max − min))
W(0)
Notes. The parameters WH and WB are expressed as the width W of the peak measured at a certain fraction x of the peak maximum, W(x). The
terms “min” and “max” refer to light-curve minimum and maximum, respectively. All values are given in phase units.
PIERBA12. Those authors classified the simulated NSs as radio quiet (RQ) or radio loud (RL) γ-ray pulsars according to
emission-geometry visibility criteria (γ-ray and radio beam orientations) and according to the γ-ray flux and radio flux sensitivity. The emission-geometry visibility criteria classify a NS as
a RL or RQ pulsar when the observer line of sight simultaneously intersects γ-ray and radio emission beams or just the γ-ray
emission beam, respectively. The γ-ray and radio flux sensitivity
criteria adopted in PIERBA12 are scaled up to three years from
one year of γ-ray LAT sensitivity and on the sensitivity of ten
radio surveys. Different γ-ray sensitivities were used for pulsars
discovered through blind search and for pulsar detected using a
timing solution known from radio observations. The radio sensitivity for RL LAT pulsars has been obtained by recomputing the
survey parameters of the ten radio surveys covering the largest
possible sky surface and for which the survey parameters were
known with high accuracy. Details about γ-ray and radio visibility computation can be found in PIERBA12.
In this paper we make use of the three-year LAT γ-ray pulsar
sensitivity and on updated radio observations of the radio loud
LAT pulsars. Since the morphological classification of the simulated light curves does not show significant variations between
the whole simulated population and its visible subsample (as we
show in Sect. 3.2), we compare the LAT morphological characteristics with the morphological characteristics of all simulated
pulsars classified as visible according to emission geometry visibility criteria only.
3. Shape classification of observed and simulated
light curves
3.1. Method
We defined a number of light-curve morphological characteristics and assigned shape classes to both simulated and observed
light curves according to the recurrence of those characteristics
in the analysed profiles. They are:
1- the number of light-curve phase windows with non-zero
emission, NE−W , defined as the number of contiguous phase
intervals in the analysed light curves that show non-zero
emission (see the second and fourth panel of Fig. 1 for
examples of two and one non-zero emission windows,
respectively);
2- the number of light-curve peaks, Pk (defined as described in
Sects. 3.1.1 and 3.1.2);
3- the number light-curve minima, Mn (defined as described in
Sects. 3.1.1 and 3.1.2);
4- the peak full width half maximum, WH , and the width at
the base of the peak, WB , for single peak light curves, both
expressed in pulse-phase units. Because of the differences in
the off-peak emission predicted by the different models and
in the observed light curves, both WH and WB values have
been optimised in the framework of each model and for observed profiles. The adopted values for γ-ray and radio light
curves, both simulated and observed, are given in Table 1. An
example of WH and WB is shown in the top panel of Fig. 1.
We used WH and WB to define a high-to-low peak width ratio, HLR = WH /WB . HLR close to 0 indicates a peak that
gets rapidly pointed while HLR close to 1 indicates a peak
with a more constant width along its vertical extension. The
parameter HLR is used for the classification of the different
kinds of single-peak light curves.
Examples of the light-curve maxima, minima, single-peak
widths, and non-zero emission windows are shown, for each
model, in Fig. 1 while the shape classes used to classify γ-ray
and radio light curves are shown in Tables A.1 and A.2, respectively. The listed morphological characteristics were computed
for observed and simulated light curves according to different
criteria.
3.1.1. LAT γ-ray and radio light curves
The γ-ray and radio observed light curves are affected by observational noise that does not allow a clear identification of
the light-curve shape morphology and must be removed before
the analysis. Both γ-ray and radio light curves of LAT pulsars
have been de-noised using a wavelet transform with an iterative
multi-scale thresholding algorithm and assuming Gaussian noise
(Starck et al. 2006). We chose Gaussian noise because the statistics of the weighted γ-ray photons follows a Gaussian distribution (PSRCAT2) and the radio light curves are well described by
Gaussian statistics. Examples of the wavelet de-noising of γ-ray
and radio LAT light curves for pulsar J0205+6449 are shown in
Fig. 2 left (red) and right (blue) panels, respectively.
The γ-ray light curve of LAT pulsars have been classified by
adopting the very same peak multiplicity assigned by PSRCAT2
while the radio light curves of LAT pulsars have been classified
on the basis of the analysis of the de-noised light curves. The
de-noised γ-ray or radio light curves also allow us to highlight
the presence of an eventual emission bridge between the peaks,
which are fundamental for the peak-separation computation (see
Sect. 4).
3.1.2. Simulated γ-ray and radio light curves
The simulated light curves are characterised by computational
fluctuations that affect the light-curve shape as real noise. The
A137, page 3 of 26
A&A 588, A137 (2016)
Original γ - ray light c urve
Original radio light c urve
0
0.2
0.4
0.6
0.8
1
0
0.2
0.4
Phase
0.2
0.4
0.6
Phase
0.8
1
γ - ray light c urve gaussian de noise d
Radi o light c urve Gaussian de noise d
0
0.6
Phase
0.8
1
0
0.2
0.4
0.6
0.8
1
Phase
Fig. 2. Examples of the wavelet Gaussian de-noising method used to
classify the LAT pulsar light curves. The de-noising of radio and γ-ray
light curves of pulsar J0205−6449 are shown on the left (red) and right
(blue) panels, respectively.
Fig. 1. Examples of light-curve classification criteria used to classify
the simulated light curves. From top to bottom: PC or radio bump light
curve; PC or radio two peaks light curve; SG double plus single peak
light curve; OG or OPC triple peak light curve.
light curves of simulated pulsars were computed using the geometrical model from Dyks et al. (2004). In the algorithm that
implements the Dyks et al. (2004) model, the number of magnetic field lines where particles are accelerated and γ-rays are
produced is an input parameter; a large magnetic field-line number yields smooth light curves with low computational fluctuations in a computational time that is potentially prohibitive,
while a low magnetic field-line density speeds up the computation but generates light curves with a critical computational
fluctuations. We chose a magnetic field line number optimised
to have a reasonable computational time and moderated computational fluctuations and we reduced the computational fluctuations by smoothing all simulated light curves using a Gaussian
filter. The smoothing consisted in computing the fast Fourier
transform (FFT) of a Gaussian function with standard deviation σν and of the simulated light curve, convolving the two
A137, page 4 of 26
FFTs, and evaluating the inverse FFT of the convolved function. The result is a light curve with all fluctuations of frequency σν smoothed. Since each model shows different computational noise, σν was optimised in the framework of each model
and for γ-ray and radio observed profiles. They are listed in
Table 1.
The smoothed light curve is then analysed by the analysis
script that detects and computes the morphological light-curve
characteristics NE−W , Pk, Mn, WH , WB , and HLR . In some cases
the main light-curve peak is comparable to the computational
fluctuations and the smoothing procedure could modify the peak
shape. Moreover, some light curves are characterised by high
off-pulse emission that does not allow a clear identification of
the light-curve emission windows NE−W (see Fig. 1). In order to
recognise computational noise maxima from light-curve peaks
and not account for the off-peak emission in the number of
emission-windows NE−W , we defined two threshold criteria. The
first criterion consists in placing at zero all light-curve intensities lower than a threshold value T h0 , usually few percent of
the absolute light-curve maximum, which is different for each
model. The second criterion consist in defining a maximum-tominimum detection threshold T hmx−mn so that a light-curve maximum is classified as a peak only if the difference between its intensity and the intensity of the previous minimum is larger than
the threshold value T hmx−mn , which is different for each model
(Fig. 1 third panel). The definition of T h0 and T hmx−mn for γ-ray
and radio light curves is given in Table 1.
The association between the morphological parameters
NE−W , Pk, Mn, WH , WB , and HLR and the γ-ray and radio shape
classes are shown in Tables 2 and 3, respectively while templates of the correspondent light-curve shapes in the framework
of each model are shown in Tables A.1 and A.2. If the number of light-curve phase windows, the number of peaks, and the
number of minima, (NE−W , Pk, Mn), detected by the analysis
script in a light curve do not correspond to any shape class listed
in Tables 2 and 3, the light curve is tagged as unclassified and
the smoothing procedure is repeated by increasing σν of a δσν .
The re-smoothing of the analysed light curve is a procedure that
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
Table 2. Assignation of the γ-ray light-curve shape classes according to a different set of light-curve morphological characteristics for PC and SG
simulated profiles (top table) and OG/OPC simulated profiles and γ-ray LAT profiles (bottom table).
PC
SG
Shape class
(NE−W , Pk, Mn)
Peak width condition
(NE−W , Pk, Mn)
1- Bump
2- Sharp
3- Shoulder
4- Two
5- Double
6- Double+Single
7- Triple/Three
8- Two double
(1, 1, 0)
(1, 1, 0)
none
(1, 2, 0)
(1, 2, 1)
(2, 3, 1)
none
(2, 4, 2)
WB ≥ 0.40 & HLR ≥ 0.50
WB < 0.40 & HLR < 0.50
/
/
/
/
/
/
OG/OPC
(1, 1, 0/1/2)
(1, 1, 0/1/2)
(1, 1, 0/1/2)
none
(1, 2, 2/3)
(1, 2, 3)
none
(1, 4, 4)
Shape class
(NE−W , Pk, Mn)
Peak width condition
(NE−W , Pk, Mn)
Peak width condition
1- Bump
2- Sharp
3- Shoulder
4- Two
5- Double
6- Double+Single
7- Triple
8- Two double
(1, 1, 0)
(1, 1, 0)
(1, 1, 0)
(2, 2, 0)
(1, 2, 0/1/2)
none
(1, 3, 1/2)
(1, 4, 3)
WB ≥ 0.20 & HLR ≥ 0.45
WB < 0.10 & HLR < 0.45
WB ≥ 0.10 & HLR < 0.45
/
/
/
/
/
(1, 1, x)
(1, 1, x)
(1, 1, x)
(2, 2, x)
(1, 2, x)
(2, 3, x)
(1/3, 3, x)
(x, 4, x)
WB ≥ 0.25 & HLR ≥ 0.40
WB ≥ 0.25 & HLR ≥ 0.40
WB ≥ 0.25 & HLR < 0.40
/
/
/
/
/
Peak width condition
WB ≥ 0.35 & HLR ≥ 0.50
WB < 0.35
WB ≥ 0.35 & HLR ≥ 0.42
/
/
/
/
/
γ-ray LAT
Notes. “None” in correspondence of a shape class for a particular model, indicates that this light-curve shape is not observed in the framework of
this model. The “peak width condition” was uniquely used to discriminate between “Bump” and “Sharp” single-peak light curves. All values are
given in phase units. Light-curve shape templates for the listed shape classes are shown in Table A.1.
Table 3. Assignation of the radio light-curve shape classes according
to a different set of light-curve morphological parameters for radio core
plus cone simulated light curves and for observed radio light curves.
Radio core plus cone / Radio LAT
Shape class
(NE−W , Pk, Mn)
Peak width condition
1- Bump
2- Sharp
3- Two
4- Double
5- Double+Single
6- Two double
7- Triple
8- Three
9- Two triple
(1, 1, 0/1)
(1, 1, 0/1)
(2, 2, 0/1)
(1, 2, 1)
(2, 3, 1)
(2, 4, 2)/(4, 4, 0)
(1, 3, 2)
(3, 3, 0)
(2, 6, 4)
WB ≥ 0.25 and HLR ≥ 0.50
WB < 0.25 and HLR < 0.50
/
/
/
/
/
/
/
Notes. The “peak width condition” was uniquely used to discriminate
between “Bump” and “Sharp” single-peak light curves. All width values
are given in phase units. Light-curve shape templates for the listed shape
classes are shown in Table A.2.
is iterated by increasing σν of a δσν at each iteration up to a sure
identification of the analysed profile. The increasing smoothing
factors δσν were optimised in the framework of each model and
are listed in Table 1.
We built the γ-ray and radio light-curve shape classifications
described in Tables 2 and 3 by merging all shape classes obtained in the framework of all observed and simulated γ-ray
light curves and observed and simulated radio light curves,
respectively. This explains why some shape classes are not observed in the framework of a particular model.
3.2. Shape classification and multiplicity: comparison
of model and observations
Figure 3 shows the recurrence of γ-ray and radio shape classes
described in Tables A.1 and A.2 respectively, for the whole simulated population synthesised by PIERBA12 and its visible subsample. The recurrence of each shape class does not change
considerably from the whole population to its visible subsample
showing that there are no important selection effects due to lightcurve shapes. The most pronounced discrepancies are observed
within the PC model where the visible subsample shows a lack of
sharp peaks (shape class 2) and an excess of double peaks (shape
class 4) and in the SG model where the double plus single peak
(shape class 8) is less recurrent among visible objects while all
other classes are more populated in the whole population. This
is confirmed by KS statistics between total and visible subsample distributions of 0.19, 0.21, 0.05, 0.10, and 0.03 for PC, SG,
OG, OPC, and radio models, respectively, with the KS statistics
ranging from 0 for total agreement to 1 for total disagreement
(see Appendix E for details). The consistency between the shape
class recurrences in the whole simulated population and its visible subsample allows us to compare the collective properties of
the whole simulated population by Pierbattista et al. (2012) with
the same properties of the observed pulsar population.
We define the light-curve-peak multiplicity as the number
of peaks detected in the radio or γ-ray light curve according
to the method described in Sect. 3. The γ-ray and radio light
curves multiplicities are associated with each shape class as
indicated in Tables A.1 and A.2, respectively. Figure 4 compares the observed and simulated peak multiplicity distributions
for the implemented γ-ray emission geometries. The statistical
agreement between observed and simulated γ-ray peak multiplicity distributions was computed by performing the twosample Kolmogorov-Smirnov (2KS). Table 4 list the 2KS test
results and suggests that the outer magnetosphere models, OG
A137, page 5 of 26
A&A 588, A137 (2016)
Table 4. Two-sample Kolmogorov-Smirnov statistics (2KS) and relative pvalue between observed and simulated one-dimensional distributions
shown in Figs. 4 to 6 for each model.
Fig. 4: γ-ray peak multiplicity
Fig. 5-left: Δγ distribution
Fig. 5-right: ΔRadio distribution
Fig. 6: δ distribution
D
PC
pvalue [%]
D
SG
pvalue [%]
D
OG
pvalue [%]
D
0.66
0.42
0.19
0.81
1e-28
2e-8
57
1e-5
0.51
0.18
0.21
0.50
7e-16
2.5
46
2e-7
0.32
0.54
0.52
0.39
2e-5
5e-15
3e-2
4e-4
0.12
0.29
0.37
0.22
OPC
pvalue [%]
23.3
3e-3
2.6
2.7
Notes. The 2KS statistics D ranges between 0 and 1 for distributions showing total agreement and total disagreement, respectively. The pvalue
is the probability to obtain the observed D value under the assumption that the two distributions are obtained from the same distribution (null
hypothesis). This is equivalent to rejecting the null hypothesis at a confidence level of 100-(pvalue )%. The 2KS test is described in Section E. The
D and pvalue parameters relative to the first and second most consistent distributions are highlighted in dark grey and light grey cells, respectively.
PC γ -ray pulsars
SG γ -ray pulsars
40
60
30
40
20
20
10
0
0
SG γ -ray pulsars
70
80
% of the total
Pulsars %
PC γ -ray pulsars
80
60
50
60
40
40
30
20
20
10
0
OG γ -ray pulsars
OPC γ -ray pulsars
80
60
% of the total
Pulsars %
60
30
40
10
20
0
0
1
2
3
4
5
γ -ray classes
6
7
8
OPC γ -ray pulsars
70
40
20
0
OG γ -ray pulsars
70
60
50
50
40
40
30
30
20
20
10
10
0
1
2
3
4
5
γ -ray classes
6
7
8
Radio pulsars
Whole population
Visible population
40
35
0
1
1s
2
3
γ -ray peak multiplicity
4
1
1s
2
3
γ -ray peak multiplicity
4
Fig. 4. Recurrence of γ-ray peak multiplicity for the simulated (grey)
and LAT (red) γ-ray pulsar populations and each model. 1s refers to
one peak with shoulder class 3 in Table A.1.
Pulsars %
30
25
20
15
10
5
0
1
2
3
4
5
6
7
8
9
Radio classes
Fig. 3. Recurrence of the shape classes described in Tables 2 and 3 and
shown in Tables A.1 and A.2 for the γ-ray pulsars (top panels), radio
pulsars (bottom panel), and each model for the whole simulated pulsar
population and its visible subsample, as synthesised by Pierbattista et al.
(2012).
and OPC, best explain the observations, with the OPC prediction showing the highest agreement with observations. The SG
model predicts too many high-multiplicity profiles and too few
double-peaked profiles. The PC model completely fails to explain the observed multiplicities.
The comparison of simulated and observed multiplicities of
γ-ray-selected radio profiles is shown in Fig. C.1. The different γ-ray model visibility criteria do not considerably affect the
shape of the radio multiplicity distribution of simulated RL pulsar and do not allow us to discriminate the γ-ray model visibility
that best explains the observations. The 2KS test results listed in
Table C.1 show that all models poorly explain the observations
A137, page 6 of 26
with the SG model prediction rejected at a lower confidence level
(CL). One must note that even though the radio model is unique,
the RL pulsars subsample changes within each γ-ray model as a
function of the γ-ray visibility so the collective radio properties
change within each γ-ray geometry. Hereafter we refer to each
model RL objects as “γ-ray selected radio pulsars”.
We studied how the γ-ray peak multiplicity changes for the
RL and RQ subsamples of observed and simulated populations.
Figure C.2 compares the recurrence of the γ-ray light-curve multiplicities in the framework of each model for RQ and RL objects
in the top and bottom panels, respectively. The γ-ray light-curve
multiplicity of observed objects shows an increase of the singlepeak light curves going from RQ to RL objects. None of the
tested emission geometries manages to reproduce the observed
behaviour of RQ and RL subsamples: from RQ to RL objects,
the PC shows an increase of peak multiplicity two, while SG
and OPC do not show significant changes in the predicted RQ
and RL distributions. The OG model peak multiplicity changes
with opposite trend with respect to observations: From RQ to
RL objects, we observe an increase of the fraction of high-peak
multiplicities 3 and 4 and a decrease of peak multiplicity 1. The
2KS tests between simulated and observed distributions shown
in Table C.1 suggest that for both RQ and RL γ-ray peak multiplicity distributions the OPC model shows the larger statistical
agreement with observations, especially for RQ objects where
the agreement reaches the 99.995%.
In order to evaluate whether a model explains the observed
shape recurrence, the peak multiplicity is a more objective indicator than the shape classes. The number of peaks in the light
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
curve is more easily recognisable and less biased by assumptions
made to classify observed and simulated profiles. Moreover the
peak multiplicity was used instead of shape classes to compare
model prediction and simulations by other studies, (e.g. Watters
et al. 2009). However, the shape classification defined in this paper enables a comparison of the observed and simulated light
curves taking more light-curve details and degenerated classes
into account, for example the multiplicity class number 2 is split
in two shape classes, double peak and two peaks, which recur
differently within the simulated and observed samples. A comparisons of the recurrence of observed and simulated γ-ray and
radio shape classes described in Tables A.1 and A.2, respectively, are given in Fig. C.3, while the statical agreement between
simulated and observed distributions is given in Table C.1. All
models poorly describe the observation with SG and OPC showing larger statistical agreements with the data.
4. Measure of the light-curve peak characteristics
We use the γ-ray and radio light curves classification described
in Sect. 3 to fit observed and simulated light curves with a number of Gaussian and/or Lorentzian functions equal to the lightcurve peak multiplicity. We tested Gaussian and Lorentzian distributions as possible shapes that best describe the simulated and
observed light-curve peaks and we obtained that, in the majority of the cases, to fit the simulated peaks with Gaussian distributions gives a lower average χ2 value with respect to the
Lorentzian fit. The Lorentzian distribution best explains some
shape-class peaks in the framework of SG, OG, OPC and γ-ray
observed light curves. The fit free parameters are: Gaussian
standard deviation or Lorentzian γ parameter (indicator of the
peak width); the function amplitude (indicator of the peak intensity), the function mean value (indicator of the peak position), and an additive constant to account for the off-peak contribution. For each radio and γ-ray light curve we obtained the
peak phases, FWHMs, intensities, as well as the peak separation in light curves with peak multiplicity larger than one. To
optimise the fitting procedure and speed up the computation, the
fit-parameter intervals were optimised in the framework of each
γ-ray and radio geometry. For instance, the PC light curves show
sharper peaks best fitted by low standard deviation Gaussians
distributions while the SG peaks are wider and often best explained by broader Lorentzian distributions. Reference values of
the Gaussian standard deviation or Lorentzian γ parameters as
indicators of the pulse width are shown, in the framework of
each model, in Table 1.
Each shape-class light curve has been fit in the framework of
each model with a number of Gaussian and/or Lorentzian functions as follows:
– PC, radio core plus cone, and LAT radio light curves
All PC, radio core plus cone, LAT radio light curves are fitted
with a combination of Gaussian functions equal to the lightcurve peak multiplicity.
– SG
The majority of the SG light curves shown to be best described by a combination of Lorentzian functions with just
the bump peak (shape class 1) and some double peaks
(shapes class 2) best described by Gaussian functions. We
noticed that the SG double-peak shapes class 5, enables
two main morphologies: the first one with smooth and
bumpy peaks and the second one with more sharp and spiky
peaks. Because of this shape dichotomy, the SG double peak
(shapes class 5) is fit with both Lorentzian and Gaussian
functions and the fit solution with the lower χ2 is considered.
– OG and OPC
OG and OPC light curves from classes 1 to 3 and 5 to 9
are best described by a single Gaussian or by a combination
of Gaussian functions while the two peaks profiles (shape
class 4) are fitted with Lorentzian functions. The different
prescription for the width of the emission gap adopted by the
OPC model (Romani & Watters 2010; Watters et al. 2009)
allows for wider profiles for all shape classes: we observe
OPC bump structures that are absent in the OG model and,
for all high-peak multiplicity shape classes, the OPC lightcurve peak appears wider than the OG peak.
– LAT γ-ray light curves
The majority of the LAT γ-ray light curves is best described by a combination of Gaussian functions. The only
case where a Lorentzian function best describes the observed
peak shape is the shoulder (shape class 3). The shoulder
structure has been fitted with two Lorentzian functions: one
to account for the low plateau skewed on the left-hand side
of the peak and the second to fit the main peak.
4.1. γ-ray and radio peak separation
We estimate peak separation for all light curves with a peak
multiplicity that is larger than one, both observed and simulated. When the light-curve peak multiplicity is equal to two, the
peak separation is given by measuring the phase interval showing bridge emission. In light curves with peak multiplicity larger
than two, Δγ was computed according to the following criteria:
γ-ray light curves
– Class 6: double plus single
Peak separation computed between the single peak and
barycentre of the double peak.
– Class 7: triple
Peak separation computed between the two extreme peaks
(only in OG and OPC cases).
– Class 8: two double
Peak separation computed between the barycentres of the
two double peaks.
Radio light curves
– Class 5: double plus single
Peak separation computed between the single peak and
barycentre of the double peak.
– Class 6: two double
Peak separation computed between each double peak
barycentre.
– Class 7: triple
Peak separation computed between the two extreme peaks.
– Class 8 and 9: three peaks and more than four peaks
Peak separation computed between the two highest
peaks.
Figure 5 compares the γ-ray peak-separation distribution Δγ and
the radio peak-separation distribution, ΔRadio , for γ-ray-selected
objects of all implemented models and LAT pulsars in the left
and right panels, respectively. The observed Δγ distribution
shown in Fig. 5 left panel, ranges in the interval 0.1 Δγ 0.55
and shows two peaks at Δγ ∼ 0.2 and Δγ ∼ 0.5. None of the proposed emission geometries manages to explain the observations,
but the SG and OPC models predict the observed distribution at
A137, page 7 of 26
A&A 588, A137 (2016)
PC γ -ray pulsars
SG γ -ray pulsars
% of the total
% of the total
PC γ -ray pulsars
25
50
20
40
30
15
20
10
10
5
0
0
40
30
30
20
20
10
10
0
OG γ -ray pulsars
0
OPC γ -ray pulsars
OG γ -ray pulsars
35
OPC γ -ray pulsars
60
20
30
25
40
% of the total
% of the total
SG γ -ray pulsars
40
15
20
10
15
10
5
0
0
0.2
0.4
Δγ
0.6
0.8
30
40
30
20
20
10
10
5
0
50
0
0
0.2
0.4
Δγ
0.6
0.8
0
0
0.1
0.2
0.3
Δ Radio
0.4
0.5
0.6
0
0.1
0.2
0.3
Δ Radio
0.4
0.5
0.6
Fig. 5. γ-ray and radio peak separation for simulated samples (grey) and LAT sample (red) and each model in the left and right panels, respectively.
high and low Δγ values, respectively; the 2KS test results listed
in Table 4 shows that SG and OPC models explain the observations with the highest and second highest statistical significance,
respectively. The PC model predicts Δγ just at low and high values: Δγ < 0.1 are generated by each magnetic pole hollow cone
while Δγ ∼ 0.5 are generated by the two emission cones from
each magnetic pole separated by 0.5 in phases that start to be
visible for high α and ζ (see Fig. B.1). The OG distribution is
antithetic to the observed distribution. This distribution shows a
maximum at Δγ ∼ 0.2 which is consistent with the observations
but goes to 0 at Δγ = 0.5 where the observed distribution shows
its absolute maximum. The SG model explains the observations
just for Δγ > 0.4, exhibits one peak at 0.35, which does not
show up in the data, and clearly underestimates the observations
for Δγ < 0.25. The OPC Δγ distribution shows a good consistency with observations for Δγ < 0.25 but completely fails to explain the observations at high Δγ . We note that SG and OPC are
complementary in how they explain the observed Δγ distribution
and how the different prescription for the gap width computation
in the OPC model affects the peak-separation distribution. The
OPC prescription for the gap width allows for wider light-curve
double peaks since the gap widths are generally smaller than
those of the OPC and this is evident by comparing the OG and
OPC peak-separation distributions; the OG distribution peaks at
0.2 and quickly goes to zero at Δγ = 0.5, while the OPC distribution shows a broad peak in the interval 0.2 < Δγ < 0.4 and slowly
decrease to a non-zero minimum at Δγ = 0.5. This partially
solves the complete lack of Δγ = 0.5 in the OG model and allows
the OPC to explain, even poorly, the observed trend. However
the vast majority of observed Δγ = 0.5 objects suggests that a
SG-like two-pole caustic emission geometry is necessary to explain the wide separation of the observed γ-ray peaks, which, on
the other hand, does not explain objects with Δγ < 0.25.
The γ-ray peak-separation distribution was also studied in
the framework of PSRCAT2. Those authors performed a γ-ray
light-curve analysis with the purpose to give the most precise
possible measures of the peak positions and separations. Even
though the purpose of our light-curve analysis is not to give precise measurements of peak positions and separations but to measure observed and simulated light curves under the same criteria
to be compared, our Δγ distribution for LAT ordinary pulsars is
consistent with the distribution obtained by PSRCAT2.
The observed radio peak-separation distribution shown in the
right panel of Fig. 5 ranges in two intervals, 0 < ΔRadio < 0.15,
where it peaks at ΔRadio ∼ 0.05, and 0.4 ΔRadio 0.55. A
comparison of the ΔRadio observed and simulated distribution in
A137, page 8 of 26
the framework of each model shows that all of the proposed
emission geometries explain the observed trend for low peak
separation ΔRadio < 0.15 while for ΔRadio > 0.4 all models over
predict the observations. This may be due to a radio cone used
in our simulation that is too large and that increases the probability that the line of sight intersects the emission from both
poles. The 2KS tests shown in Table 4 suggest that all models
explain well the observations with the two-pole emission geometries, PC and SG, best explaining the LAT distribution since they
predict the observed proportion. In analogy to the PC geometry,
the ΔRadio ∼ 0.1 are generated by double-peak structures typical
of each radio emission beam, while ΔRadio ∼ 0.5 are given by the
distances of the two emission cones from each magnetic pole for
high α and ζ (see Fig. B.1).
4.2. Radio lag
We consider the radio and γ-ray light curves of the same pulsar,
both coherent in phase with the actual pulsar rotational phase.
The radio lag is then defined as the phase lag between a radio
fiducial phase and the following γ-ray peak phase. The radio lag
is considered as a tracker of the pulsar magnetosphere structure.
The radio lag constrains the relative positions of γ-ray and radio emission regions in the pulsar magnetosphere and can be
used to discriminate the proposed γ-ray and radio emission geometries that best explain the observations. While it is relatively
easy to produce radio and γ-ray light curves that are both coherent in phase with the pulsar rotation through timing techniques,
the definition of the radio fiducial phase is more controversial.
The commonly accepted definition for the radio fiducial phase is
the phase of the radio peak following the pulsar magnetic pole
that should be identified, case by case, by analysing the lightcurve shape. However, this definition is strongly dependent on
the light-curves quality, which might not be good enough to enable a robust identification of the magnetic pole, and might be biased by the radio model used to predict the magnetic pole phase
in the observed light curve.
In PSRCAT2 the problem of finding solid criteria to assign
a radio fiducial phase to observed light curves was solved by increasing the quality of the radio light curve with deeper radio
observations of the analysed objects and by defining a series of
morphological criteria. When the light curve shows symmetric
structures (double or higher multiplicity peaks) the pulsar fiducial phase is associated with the peak barycentre. When radio
and γ-ray peaks are aligned (e.g. Crab pulsar), the radio emission cannot be explained by a conical beam generated above the
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
PC γ -ray and radio pulsars
% of the total
40
SG γ -ray and radio pulsars
25
20
30
15
20
10
10
5
0
0
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
35
% of the total
magnetic poles but rather by caustic emission, as proposed for
the Crab pulsar by Venter et al. (2012). In these cases the fiducial phase is associated with the phase of the radio precursor,
which is a small light-curve feature.
The problem of finding a robust definition for the radio fiducial phase becomes critical when one compares simulations with
observations. The magnetic-pole phase for simulated objects is
known with high precision, it is independent of the assumed radio and γ-ray emission geometries, and is not affected by the
light-curve quality. It is thus easy to identify the phase of the
first peak after the magnetic pole. Moreover, the cone plus core
radio emission geometry we adopt does not predict caustic radio
emission and no aligned radio and γ-ray peaks are possible. The
obvious consequence is that observed and simulated radio-lag
measurements are not completely consistent and the conclusions
drawn from their comparison loses scientific reliability. The radio lag for simulated objects has been computed as the phase
separation between the first radio peak detected by the analysis algorithm, the fiducial phase, and the phase of the following γ-ray peak. Both γ-ray and radio simulated light curves have
been generated within the phase interval −0.5 to 0.5 with the pulsar fiducial plane (the plane containing magnetic and rotational
axes) intersecting the line of sight at phases 0 and ±0.5. Because
of that, the fiducial phase was prevalently found at phase 0 and,
for large α and ζ values, at phase −0.5 (see Fig. B.1). In twopole γ-ray emission geometries, PC and SG, each magnetic pole
shines both in radio and in γ-rays, and the radio lag is always
measured between the fiducial phase and the γ-ray beam coming
from the same magnetic hemisphere. In one-pole γ-ray emission
geometries there is just one magnetic hemisphere that shines in
γ-rays and when, for large α and ζ, radio emission from both
magnetic poles starts to be visible in the light curve, cases of
radio lag larger than 0.5 are possible; e.g. when radio emission
beam starts shining at phase −0.5, the algorithm sets the fiducial
phase equal to −0.5 and measures the radio lag as its distance
from the γ-ray peak generated in the opposite magnetic hemisphere. All δ > 0.5 for OG and OPC models were subtracted
of 0.5 to measure the radio lag as the phase lag between fiducial
phase and γ-ray peak in the framework of the same magnetic
hemisphere.
We use the radio and γ-ray light curves published in
PSRCAT2 to estimate the radio lags of RL LAT pulsars as the
phase separation between the fiducial phase and the following
γ-ray peak. We define the fiducial phase as the phase of the first
peak, which appears in the radio light curve as detected from
the algorithm described in Sects. 3 and 4, to measure the radio
lag for RL LAT pulsars consistent with the simulated sample.
When two peaks separated by 0.5 in phase are visible in the radio light curve, we assign the fiducial phase to the radio peak for
which the radio lag is smaller than 0.5. In this way we avoid the
assumption about the magnetic pole position and the radio precursor that is not modelled in the framework of the implemented
radio-beam emission geometry.
Figure 6 compares the radio lag δ distribution of simulated
and observed RL pulsars in the framework of each model. A
comparison between our δ distribution and the distribution obtained by PSRCAT2 shows a total consistency both in the range
of values and in the proportions. Our δ distribution ranges in
the interval 0 < δ < 0.5, and raises steeply up to its maximum at δ = 0.15 and shows a stable flat trend in the interval
0.2 < δ < 0.5. None of the tested models manages to explain
the observed distributions over all the range of observed δ but
the OPC shows higher agreement with the data. The 2KS test
statistics given in Table 4 shows that the OPC model explains
20
30
25
15
20
10
15
10
5
5
0
0
0
0.1
0.2
0.3
δ
0.4
0.5
0.6
0
0.1
0.2
0.3
δ
0.4
0.5
0.6
Fig. 6. Radio-lag distribution for simulated samples (grey) and LAT pulsars (red) and each model.
the observations with a CL at least four orders of magnitude
larger than the other models and that the OG predictions explain
the observations with the second highest CL. Both PC and SG
models predict emission beams most of the time overlapping the
radio emission beam, with the PC emission beam tightly matching the radio beam both in size and pulse phase (see Fig. B.1).
This generates the excess of δ < 0.1 predicted by both PC and
SG emission geometry with the SG geometry predicting larger
δ values because of its wider emission beam. The OG model
completely fails in predicting the observed distribution both in
shape and proportions. The OPC is the model that best predicts
the observed δ range and proportions. This model predicts a peak
that is too broad in the range 0.15 < δ < 0.25 which partly overlap the observed peak but underestimates objects in the range
0.35 < δ < 0.5. Of particular interest is the difference between
OG and OPC distributions. As also shown in Fig. 5, the different prescription for the gap width adopted by the OPC enables
broader light-curve peaks that occur closer, in phase, to the radio peak, thereby decreasing the radio lag. Under the assumption that the radio model used in this simulation is correct and
that the adopted VRD magnetic field geometry is correct, this
picture supports an outer magnetosphere location of the emission gap, points to the need to model broader light-curve γ-ray
peaks, and highlights the importance of the gap width and B field
structure choice in the light-curve modelling.
One has to note that the model radio lags obtained in this
study are strongly dependent on the assumed VRD magnetic
field geometry. Kalapotharakos et al. (2012) compute the γ-ray
light curves in a dissipative magnetosphere (DM) with conductivity ranging from 0 to ∞, corresponding to VRD magnetic field
geometry and to a force-free electrodynamics (FFE) magnetic
field geometry, respectively. Those authors conclude that VRD
magnetospheres shows the smallest radio to γ-ray lags while
more realistic DMs predict average larger lags (Kalapotharakos
et al. 2012). All γ-ray model radio-lag distributions computed
in this study would be shifted towards higher values if computed in the framework of a DM, and this would solve the excess of low radio-lag values predicted by the PC and SG models. Particularly interesting is the case of the SG model that
shows the correct shape of the observed radio-lag distribution
but shifted towards values that are too low; the implementation
of the SG model within a DM may considerably improve the SG
prediction.
A137, page 9 of 26
A&A 588, A137 (2016)
4.3. Non-observable pulsar parameters: Δγ , ΔRadio , and δ
as a function of fΩ , α, ζ
0.8
PC γ -ray pulsars
SG γ -ray pulsars
OG γ -ray pulsars
OPC γ -ray pulsars
0.7
0.6
A137, page 10 of 26
Δγ
0.5
0.4
0.3
0.2
0.1
0
0.8
0.7
0.6
Δγ
0.5
0.4
0.3
0.2
0.1
0
0
0.5
1
1.5
2
0
0.5
1
1.5
2
fΩ
fΩ
PC γ -ray and radio pulsars
SG γ -ray and radio pulsars
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
0.6
0.5
Δ r adio
0.4
0.3
0.2
0.1
0
0.6
0.5
0.4
Δ r adio
We compare simulated and estimated correlations between Δγ ,
ΔRadio , and δ and non-observable pulsar parameters, namely
beaming factor fΩ , magnetic obliquity α, and observer line-ofsight ζ. The simulated distributions of fΩ , α, and ζ are those
synthesised in the framework of PC, SG, OG, and OPC emission models by PIERBA12 via the emission geometry proposed
by Dyks et al. (2004). The same parameters for LAT pulsars have
been estimated from fits to the observed pulsar light curves by
PIERBA15 in the framework of the same emission models and
using the same emission geometry by Dyks et al. (2004).
The simulation by PIERBA12 gives the allowed range of values for the parameters fΩ , α, and ζ within each model. Since fΩ ,
α, and ζ for LAT pulsars were estimated by implementing the
same geometrical model used by PIERBA12 to synthesise the
simulated-population light curves, we expect a match between
estimations and simulations within the same model only if the
model manages to reproduce the variety of the observed lightcurve shapes. If the variety of observed light-curve shapes is not
explained in the framework of a particular model, the best-fit
parameters fΩ , α, and ζ do not match the interval of the most
likely values as simulated by PIERBA12 in the framework of
the same model. Because of that, a mismatch between estimated
(PIERBA15) and simulated (PIERBA12) trends in the framework of a particular model suggests the inadequacy of that model
in explaining the observations. An example of inadequacy of a
model in explaining the observed light curves is given by the
PC model and it is evident by looking at the PC panel at the
top of Fig. 8. The vast majority of the PC light curves show
sharp peaks and low off-peak emission that do not manage to
predict the variety of the LAT light-curve shapes. This implies
that the PIERBA15 fitting algorithm selects as best-fit solutions
only PC light curves with very low α and ζ angles, which are
those characterised by broader peaks and more likely to explain
the observed shapes (see Fig. B.1). As a consequence, in the
PC panel of Figs. 8 and 9, the estimated points are all grouped
at low α and ζ values and do not match the α and ζ intervals
predicted by the PC simulation. This suggests the inadequacy
of the PC model in predicting the observations. The statistical
agreement between observed/estimated and simulated distributions was quantified by computing the two-sample KolmogorovSmirnov (2KS) statistics for the two-dimensional distributions
proposed by Press et al. (1992) as described in Sect. E. The D
and pvalue values computed between each observed/estimated
and simulated distribution are listed in Table 5.
Figure 7 compares estimated and simulated distributions for
Δγ and ΔRadio as a function of the beaming factor fΩ and for each
model. The Δγ − fΩ plane is a tracker of the magnetospheric region where the γ-ray emission is generated since both Δγ and fΩ
depends on the emission-region structure. Overall none of the
distributions estimated in the framework of the proposed emission geometries matches the simulated ranges but the SG and
OG models show consistency with observations at the highest
and second highest CL, respectively, as indicated in Table 5.
Both OG and OPC models show highly dispersed distribution
and suggest that, for the bulk of the simulated distributions, Δγ
increases as fΩ increases. The OPC estimates match the simulations only for fΩ > 0.4 since they manage to reproduce the bulk
of objects centred at fΩ ∼ 0.85 and Δγ ∼0.45 while the OG estimates do not match the bulk of the simulated objects. The PC
and SG models do not predict any Δγ variation as fΩ changes
and, in both cases, the estimates do not match the bulk of the
0.3
0.2
0.1
0
0
0.5
1
fΩ
1.5
2
0
0.5
1
1.5
2
fΩ
Fig. 7. Distributions of the γ-ray and radio peak separation of simulated (grey) and observed (red) pulsars obtained, for each model, as
a function of the γ-ray beaming factor are shown in the top and bottom panels, respectively. The model distributions have been obtained
as two-dimensional number-density histograms. The dash-dotted line
highlights the minimum number density contour. The LAT measurements are given with 1σ error bar uncertainty.
simulations. In the ΔRadio − fΩ plane, all model estimates reasonably match the simulated trends and given the low number of
observed objects, the relative pvalues listed in Table 5 cannot be
used to discriminate the model that best explains the data.
Figures 8 and 9 compare estimated and simulated trends of
Δγ and ΔRadio as function of α and ζ, respectively. The parameters Δγ and ΔRadio are expected to change with increasing α and ζ
with different trends and in the framework of different emission
geometries, as shown in Fig. B.1. In the PC and radio emission
geometries, the size of the conical emission beams decrease as α
and ζ increase, therefore Δγ and ΔRadio are expected to decrease
with increasing α and ζ. This behaviour is barely visible in the
PC γ-ray separation but it is evident in the radio cases for PC and
SG-selected objects. The PC estimates populate the low α and ζ
regions of the Δγ/radio − α and Δγ/radio − ζ planes, respectively,
and do not match the major fraction of simulated points. The
SG model predicts increasing Δγ when both α and ζ increase.
As shown in Fig. B.1, the bright SG caustic widens when ζ increases for fixed α and when α increases, the caustic from the
second magnetic pole starts to be visible in the light curve and
Δγ tend to 0.5. In both Δγ −α and Δγ −ζ planes, the SG estimates
match the simulated trends. The OG and OPC models are explained in the framework of a one-pole caustic emission geometry; when α increases, no emission from the other pole shows up
at high ζ and Δγ does not tend to 0.5 as α or ζ increase. This is
less evident in the OPC model predictions where, because of the
different prescription for the gap width, wider peaks are possible
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
PC γ -ray pulsars
SG γ -ray pulsars
0.8
0.7
0.7
0.6
0.6
0.5
0.5
Δγ
Δγ
0.8
0.4
SG γ -ray pulsars
OG γ -ray pulsars
OPC γ -ray pulsars
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
0
OG γ -ray pulsars
0.8
OPC γ -ray pulsars
0.8
0.7
0.7
0.6
0.6
0.5
0.5
Δγ
Δγ
PC γ -ray pulsars
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
0
0
10
20
30
40
50
α [deg]
60
70
80
90
0
10
PC γ -ray and radio pulsars
20
30
40
50
α [deg]
60
70
80
90
0
10
SG γ -ray and radio pulsars
0.5
0.4
0.4
30
40
50
ζ [deg]
60
70
80
90
0
10
20
30
40
50
ζ [deg]
60
70
PC γ -ray and radio pulsars
SG γ -ray and radio pulsars
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
80
90
80
90
Δ radio
0.6
0.5
Δ radio
0.6
20
0.3
0.3
0.2
0.2
0.1
0.1
0
0
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
0.5
0.5
0.4
0.4
Δ r adio
0.6
Δ r adio
0.6
0.3
0.3
0.2
0.2
0.1
0.1
0
0
0
10
20
30
40
50
α [deg]
60
70
80
90
0
10
20
30
40
50
α [deg]
60
70
80
90
0
10
20
30
40
50
ζ [deg]
60
70
80
90
0
10
20
30
40
50
ζ [deg]
60
70
Fig. 8. Distributions of the γ-ray and radio peak separation of simulated (grey) and observed (red) pulsars obtained, for each model, as
a function of the magnetic obliquity are shown in the top and bottom panels, respectively. The model distributions have been obtained
as two-dimensional number-density histograms. The dash-dotted line
highlights the minimum number density contour. The LAT measurements are given with 1σ error bar uncertainty.
Fig. 9. Distributions of the γ-ray and radio peak separation of simulated (grey) and observed (red) pulsars obtained, for each model, as a
function of the observer line of sight are shown in the top and bottom panels, respectively. The model distributions have been obtained
as two-dimensional number-density histograms. The dash-dotted line
highlights the minimum number density contour. The LAT measurements are given with 1σ error bar uncertainty.
and both Δγ − α and Δγ − ζ trends show broader distributions.
Both Figs. 8 and 9 show that Δγ increases for increasing α or
ζ with the OPC trend showing higher dispersion with respect to
the OG trend as a result of the different prescription used in the
OPC to compute the width of the accelerator gap. In the plane
Δγ −α, neither OG nor OPC estimates match the bulk of the simulated distribution; in both cases, the estimates show an excess
at Δγ = 0.5, which is not predicted by the models. In the plane
Δγ − ζ, the OG estimates over predict Δγ = 0.5 and fail again in
matching the bulk of the simulated distribution, while the OPC
estimates match the simulation explaining the simulated increasing trend well. The results of the statistical test shown in Table 5
find that in the plane Δγ − α, PC and SG models predict the observations with the highest and second highest CL while in the
Δγ − ζ plane, the OG and SG models give the highest and second
highest CL predictions of the data. The best agreement obtained
between observed/estimated and simulated distributions in the
plane Δγ − α is fictitious since the distributions are totally not
overlapping and inconsistent.
In the ΔRadio −α and ΔRadio −ζ planes, as a consequence of the
radio beam shrinking with increasing α or ζ, the simulated ΔRadio
are expected to decrease with increasing α and ζ, especially at
low angles, as is evident from Fig. B.1. This trend is well visible
in the PC- and SG-selected objects with α and ζ values ranging from 0 to 90 degrees while the trend is not appreciable for
OG- and OPC-selected objects, which are mainly characterised
by high α and ζ. Radio objects selected in the framework of all
γ-ray models show an excess for α and ζ > 70◦ and Δγ = 0.5. As
in the PC case, this excess is a consequence of the radio emission
from the other pole that, for high ζ angles, starts to be visible for
high α values and allows peak separations equal to the magnetic
pole distance, 0.5. Overall, all model estimates match the simulated trends with consistent CL with PC and OG showing the
highest CL in the ΔRadio − α and ΔRadio − ζ planes, respectively
(see Table 5).
Figure 10 compares estimated and simulated trends for δ as
functions of fΩ , α, and ζ, in the top, middle, and bottom panels, respectively. In the fΩ − δ plane, the two-pole emission geometries PC and SG, do not predict any trend for δ changing
with increasing fΩ while the one-pole caustic models, OG and
OPC, predict mild decreasing δ as fΩ increases. The estimates
obtained in the framework of PC and SG emission geometries
fail to match the model prediction and do not match the bulk
of the objects modelled while OG and OPC estimates match the
bulk of the model prediction and seem to best represent the modelled behaviour showing the highest and second highest pvalue
(see Table 5).
In the α − δ and ζ − δ planes, a decreasing trend of δ as α
and ζ increase clearly shows up in the simulations for all models except for the PC. This behaviour can be explained in the
light of Fig. B.1; as α increases, the bright caustic approaches
the radio emission beam, thereby decreasing δ. The very same
A137, page 11 of 26
A&A 588, A137 (2016)
PC γ -ray and radio pulsars
SG γ -ray and radio pulsars
0.8
0.6
PC γ -ray pulsars
SG γ -ray pulsars
OG γ -ray pulsars
OPC γ -ray pulsars
0.7
0.5
0.6
0.5
δ
Δγ
0.4
0.3
0.4
0.3
0.2
0.2
0.1
0.1
0
0
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
0.8
0.6
0.7
0.6
0.4
0.5
δ
Δγ
0.5
0.3
0.4
0.3
0.2
0.2
0.1
0.1
0
0
0
0.5
1
1.5
2
0
0.5
fΩ
1
1.5
25
2
26
fΩ
PC γ -ray and radio pulsars
27
28
29
30
31
32
25
26
log10( Ė [W])
SG γ -ray and radio pulsars
0.6
0.5
0.5
0.4
0.4
28
29
30
31
32
log10( Ė [W])
SG γ -ray and radio pulsars
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
δ
Δ radio
0.6
27
PC γ -ray and radio pulsars
0.3
0.3
0.2
0.2
0.1
0.1
0
0
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
0.6
0.5
0.5
0.4
0.4
δ
Δ r adio
0.6
0.3
0.3
0.2
0.2
0.1
0.1
0
0
0
10
20
30
40
50
α [deg]
60
70
80
90
0
10
PC γ -ray and radio pulsars
20
30
40
50
α [deg]
60
70
80
25
90
26
27
28
29
30
31
32
25
26
0.6
0.5
0.5
0.4
0.4
28
29
30
31
PC γ -ray and radio pulsars
SG γ -ray and radio pulsars
32
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
δ
δ
0.6
27
log10( Ė [W])
log10( Ė [W])
SG γ -ray and radio pulsars
0.3
0.3
0.2
0.2
0.1
0.1
0
0
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
0.6
0.5
0.5
0.4
0.4
δ
δ
0.6
0.3
0.3
0.2
0.2
0.1
0.1
0
0
0
10
20
30
40
50
ζ [deg]
60
70
80
90
0
10
20
30
40
50
ζ [deg]
60
70
80
90
Fig. 10. Distributions of the radio lag of simulated (grey) and observed (red) pulsars obtained, for each model, as a function of the
γ-ray beaming factor, magnetic obliquity, and observer line of sight
are shown in the top, central, and bottom panels, respectively. The
model distributions have been obtained as two-dimensional numberdensity histograms. The dash-dotted line highlights the minimum number density contour. The LAT measurements are given with 1σ error bar
uncertainty.
behaviour is observed in each phase-plot panel when ζ increases
for fixed α values; in both SG and OG/OPC emission geometry the bright caustic gets closer to the radio emission beam
when ζ increases. The same decreasing trend should also be observed in the PC model prediction since γ-ray and radio emission
beams get closer when α increase but the trend does not show up
because of the paucity of simulated PC objects. The outer and
intermediate-high magnetosphere model estimates from OPC
and SG best match the modelled trends showing the largest pvalue
values (see Table 5.)
A137, page 12 of 26
25
26
27
28
29
log10( Ė [W])
30
31
32
25
26
27
28
29
30
31
32
log10( Ė [W])
Fig. 11. Distributions of the γ-ray peak separation, radio peak separation, and radio lag of simulated (grey) and observed (red) pulsars obtained, for each model, as a function of the spin-down power are shown
in the top, central, and bottom panels, respectively. The model distributions have been obtained as two-dimensional number-density histograms. The dash-dotted line highlights the minimum number density contour. The LAT measurements are given with 1σ error bar
uncertainty.
4.4. Observable pulsar parameters: Δγ , ΔRadio , and δ
as a function of spin-down power Ė and spin period P
We compare the simulated and observed correlations between
Δγ , ΔRadio , and δ and pulsar spin-down power Ė and spin period
P. Simulated and observed Ė values were computed as described
in PIERBA12 while the LAT pulsar spin periods have been taken
from PSRCAT2. The statistical agreements between observed
and simulated distributions are given in Table 5. Figure 11 compares observed and simulated trends of Δγ , ΔRadio , and δ as a
function of Ė in the top, middle, and bottom panels, respectively.
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
0.8
PC γ -ray pulsars
SG γ -ray pulsars
OG γ -ray pulsars
OPC γ -ray pulsars
0.7
0.6
Δγ
0.5
0.4
0.3
0.2
0.1
0
0.8
0.7
0.6
Δγ
0.5
0.4
0.3
0.2
0.1
0
−1.6
−1.4
−1.2
−1
−0.8
−0.6
−0.4
−0.2
0
−1.6
−1.4
−1.2
−1
log10(P [s])
−0.8
−0.6
−0.4
−0.2
0
log10(P [s])
PC γ -ray and radio pulsars
SG γ -ray and radio pulsars
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
0.6
0.5
Δ radio
0.4
0.3
0.2
0.1
0
0.6
0.5
Δ r adio
0.4
0.3
0.2
0.1
0
−1.6
−1.4
−1.2
−1
−0.8
−0.6
−0.4
−0.2
0
−1.6
−1.4
−1.2
−1
log10(P [s])
−0.8
−0.6
−0.4
−0.2
0
log10(P [s])
PC γ -ray and radio pulsars
SG γ -ray and radio pulsars
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
0.6
0.5
δ
0.4
0.3
0.2
0.1
0
0.6
0.5
δ
0.4
0.3
0.2
0.1
0
−1.6
−1.4
−1.2
−1
−0.8
−0.6
log10(P [s])
−0.4
−0.2
0
−1.6
−1.4
−1.2
−1
−0.8
−0.6
−0.4
−0.2
0
log10(P [s])
Fig. 12. Distributions of the γ-ray peak separation, radio peak separation, and radio lag of simulated (grey) and observed (red) pulsars obtained, for each model, as a function of the spin period are shown in the
top, central, and bottom panels, respectively. The model distributions
have been obtained as two-dimensional number-density histograms.
The dash-dotted line highlights the minimum number density contour.
The LAT measurements are given with 1σ error bar uncertainty.
The comparisons of simulated and observed trends in the framework of all models reflect the same inconsistency shown in
PIERBA12; none of the simulated emission geometries manage to explain the behaviour of the observed sample at high Ė
showing a deficiency of high-Ė objects with respect to the observations. In the plane Δγ − Ė the PC model completely fails
to explain the observations. As already shown in Fig. 5, the SG
model predicts a double Δγ distribution. The first distribution
is centred at Δγ = 0.5 and is given by the separation of the
peaks coming from each magnetic pole caustic at high α and
does not change with increasing Ė. The second distribution is
centred at Δγ = 0.35, is given by the separation of the double
peaks generated by the same magnetic-pole caustic, and shows a
mild decrease with increasing Ė due to the shrinking of the SG
beam as Ė increases. The models SG and OPC best explain the
observations with OPC best predicting the observed behaviour.
The OG model over predicts low Ė objects for Ė < 1026.5 W,
while OPC does not predict any object in the same interval. The
pvalue values listed in Table 5 shows that in most of the cases,
OPC and OG describe the observed distributions at the higher
CL with the SG predictions best explaining the observations on
plane Ė − Δγ .
The Ė − Δγ correlation was also studied in the framework of
PSRCAT2. A comparison of the LAT pulsars Ė − Δγ correlation
obtained here and in the framework of PSRCAT2 shows consistency between the two Δγ computation techniques. As a result
of more accurate Δγ measures, the Ė − Δγ correlation obtained
by PSRCAT2 shows clearer evidence of a double trend that is
similar to our SG model predictions but centred at lower Δγ values of 0.43 and 0.22, with no apparent decrease of the Δγ as Ė
increase. In the Ė − ΔRadio plane all γ-ray-selected radio pulsars
show the same double trend at constant Ė values, 0.05 and 0.5,
with the OPC-selected radio sample preferred to explain the observations. In the Ė − δ plane, all implemented emission geometries fail to explain how the radio lag changes with increasing Ė.
Both PC and SG models do not show any trend in δ changing
with increasing Ė, while the one-pole caustic models, OG and
OPC, show a mild trend in δ decreasing with increasing Ė.
Figure 12 compares observed and simulated trends of Δγ ,
ΔRadio , and δ as a function of spin period P in the top, middle,
and bottom panels, respectively. In contrast to the Ė computation, which requires assumptions on the pulsar mass, radius, and
moment of inertia, the spin period is an assumption-independent
of observed characteristics. An increase of P implies an increase
of the light-cylinder radius corresponding to a decrease of the
open magnetic field-line region. In the P−Δγ plane, the observed
sample shows a two-component distribution: one with objects
characterised by Δγ ∼ 0.5 as P increases and one showing Δγ
decreasing as P increases for Δγ 0.45. Both OG and OPC
predict the observed decreasing component as P increases but
overestimate the high P objects and fail to explain the observed
flat component at Δγ ∼ 0.5. The OPC best explains the observations since best describes the observed distribution over the entire P interval. In the PC case, the model predictions completely
fail to explain the observed decreasing trend. The increase of
the light-cylinder radius implies a shrinking of the conical emission beams and a decrease of Δγ with P. Because of the paucity
of PC simulated objects, a mild peak separation decrease is observed just for the γ-ray-selected radio simulated objects and for
P < 125 ms. The SG model predicts a constant double distribution centred at Δγ values of 0.35 and 0.5. The model explains
the constant Δγ ∼ 0.5 branch but fails to describe the observed
decreasing trend. The fact that PS/SG and OG/OPC best predict the flat and decreasing component of the observed distribution, respectively, is because of the two-pole and one-pole nature
of their emission, respectively; the two-pole emission predicts
more 0.5 separated peaks from both the poles, while in one-pole
emission, 0.5 separated peaks only occur in the shorter P light
curves. In the P−ΔRadio plane the radio objects selected in framework of each γ-ray model give the very same explanation of the
observed points and cannot be used to discriminate the model
that best explains the observations.
In the plane P − δ none of the emission geometries manages
to describe the observed trend but the OPC models seem to give a
A137, page 13 of 26
A&A 588, A137 (2016)
PC γ -ray and radio pulsars
0.8
SG γ -ray and radio pulsars
PC γ -ray and radio pulsars
SG γ -ray and radio pulsars
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
0.6
0.7
0.5
0.6
0.4
Δ r adio
Δγ
0.5
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
0
OG γ -ray and radio pulsars
0.8
OPC γ -ray and radio pulsars
0.6
0.7
0.5
0.6
0.4
Δ r adio
Δγ
0.5
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
0
0
0.1
0.2
0.3
δ
0.4
0.5
0.6
0
0.1
0.2
0.3
δ
0.4
0.5
0.6
0
0.1
0.2
0.3
δ
0.4
0.5
0.6
0
0.1
0.2
0.3
δ
0.4
0.5
0.6
Fig. 13. Distributions of the γ-ray and radio peak separation of simulated (grey) and observed (red) pulsars obtained, for each model, as a function
of radio lag are shown in the left and right panels, respectively. The model distributions have been obtained as two-dimensional number-density
histograms. The dash-dotted line highlights the minimum number density contour. The LAT measurements are given with 1σ error bar uncertainty.
better description of the observations. The shrinking of the open
field-line region as a consequence of the increase of P implies
a simultaneous shrinking of both radio and γ-ray beams in the
framework of each model. In the SG case (and mildly in the PC),
we observe a decreasing trend in δ as P increases, which could
be explained by a faster shrinking of the radio emission beam
with respect to the PC beam and as a consequence of the SG
caustic approaching the radio beam as P increases, respectively.
The OG and OPC model predictions are not able to explain the
observations; both OG and OPC simulations are characterised
by a mild increase of δ as P increases, which is not present in
the observed distribution. The statistical significance with which
the modelled distributions for Δγ , ΔRadio , and δ as a function of
P explain the observations partly mimic that obtained for the
same parameters as a function of Ė. The OPC model explains
the observed distributions at the higher CL in the planes Δγ −
P and δ − P, while the PC model predictions best explain the
observations on the plane P − ΔRadio .
4.5. γ-ray and radio peak separation as a function
of the radio lag
The relation Δγ vs. δ is particularly important in the study of
the pulsar emission geometry and magnetosphere structure since
it correlates with information on the magnetospheric regions
where γ-ray and radio emission are generated: Δγ is a function
of the pulsar gap width structure while δ constrains the relative
position of γ-ray and radio emission regions. The Δγ vs. δ correlation was first studied by Romani & Yadigaroglu (1995). By
analysing the light curves of six radio pulsars detected in γ-rays
from the Compton gamma-ray observatory (CGRO), those authors concluded that Δγ shows a decreasing trend of increasing
δ. With the sizeable increase of the number of γ-ray and radio
active pulsars because of the advent of the Fermi satellite, it was
finally possible to give more precise estimates of the δ − Δγ relation. Watters et al. (2009) performed a first measurement of
δ and Δγ for six RL LAT pulsars confirming the trend Δγ decreasing with increasing δ. The trend was further confirmed in
the framework of PSRCAT1 on a sample of 17 RL pulsars and
by independent analyses on the same pulsar sample performed
by Pierbattista (2010). The last and more accurate δ and Δγ
measurements were obtained in the framework of PSRCAT2;
these measurements concern 32 pulsars and confirm the previous findings of Δγ decreasing with increasing δ. In all the above
A137, page 14 of 26
mentioned cases, the radio lag δ was computed as the distance
between the fiducial phase computed according to PSRCAT2
criteria (estimated position of the magnetic pole) and the phase
of the followingγ-ray peak. Since radio and γ-ray peak refer to
the same magnetic pole, no δ > 0.5 is allowed.
We compared simulated and observed correlations Δγ vs. δ
and ΔRadio vs. δ. For both simulated and observed objects, Δγ
and ΔRadio have been computed as described in Sect. 4.1 while
the computation technique of δ is described in Sect. 4.2. We studied how the radio peak separation changes with increasing δ to
constrain the structure of the radio emission region against the
models.
Figure 13 compares observed and simulated trends for Δγ
vs. δ and ΔRadio vs. δ in the left and right panels, respectively.
In the δ − Δγ plane, the LAT objects follow a clearly decreasing
trend of Δγ when δ increases. The PC model fails completely
to explain the observations, predicting a distribution in total disagreement with the observations. Points are only predicted for
small δ and Δγ values and seem to show a trend of increasing Δγ
with increasing δ. The same increasing trend and inconsistency
with the observations was obtained by Pierbattista (2010) comparing simulations and PSRCAT1 observations . The SG model
does not explain the observed decreasing trend, predicts observations mainly at low δ and intermediate Δγ and shows a mild increasing trend of Δγ with increasing δ. Comparisons of SG simulations against observations were performed by Watters et al.
(2009) and by Pierbattista (2010) on six γ-ray pulsars (see references in Watters et al. 2009) and on the pulsars from PSRCAT1,
respectively. The two studies found results consistent with each
other and also conclude that the SG model does not explain the
observed δ − Δγ trend. The one-pole emission geometry manages to explain the LAT findings successfully. Both the OG and
OPC models manage to reproduce the shape of the observed distribution with the OPC showing higher agreement with observations. The pvalue shown in Table 5 shows that, not including the
PC model that shows a statistics that is too low to be correctly
compared with observations, the OPC and OG are the models
that explain the observed Δγ as a function of δ with the highest
and second highest CL, respectively. Comparisons of OG simulations against observations were also performed by Watters
et al. (2009) and Pierbattista (2010) on six γ-ray pulsars and
on the pulsars from PSRCAT1, respectively (see Watters et al.
(2009) and PSRCAT1 for references about the LAT pulsars). In
both studies, the OG model predicts a decreasing trend of Δγ as
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
Peak Multiplicity
0
PC
Peak separation
0.0-0.1
0.1-0.2
0.2-0.3
0.3-0.4
>0.4
ONE
SHOULDER
TWO
THREE
FOUR
10
20
ζ [deg]
30
40
50
60
70
80
90
0
SG
10
20
30
ζ [deg]
δ increases consistent with the current findings. It is proper to
stress again (see Sect. 4.2) that the results obtained in this study
are strongly dependent on the VRD magnetic field geometry assumed. The assumption of a more realistic DM geometry would
imply larger radio to γ-ray lags predicted by all implemented
γ-ray models (Kalapotharakos et al. 2012). This would shift all
observed distributions towards higher δ and would probably imply an improvement of the SG model predictions in explaining
the observations.
In the plane δ − ΔRadio the LAT objects follow a clear flat
trend with ΔRadio stable at 0.05 with increasing δ. Both PC and
SG models predict an increase of ΔRadio as δ increases, which is
not observed in the LAT data. The simultaneous increase of Δγ
and ΔRadio as δ increases could be explained by the radio emission beam size, which increases at a higher rate than the PC and
SG beam sizes, thereby increasing the distance between the first
radio peak and the first γ-ray peak. Both outer magnetosphere
models, OG and OPC, correctly predict the observed behaviour
with the OPC explaining the observed range of δ values. In the
δ − ΔRadio plane, the pvalue shows that the OG and PC are the
models that describe the observations with the highest and second highest CL, respectively.
40
50
60
5. The α − ζ plane
70
80
90
0
OG
10
20
ζ [deg]
30
40
50
60
70
80
90
0
OPC
10
20
30
ζ [deg]
Figure 14 compares light-curve multiplicity and peak separation as a function of α and ζ for modelled and observed γ-ray
light curves. The α and ζ of the LAT pulsars are taken from
PIERBA15 while the α and ζ of the simulated objects are
those synthesised by PIERBA12. Parameters α and ζ are nonobservable pulsar characteristics and, analogous to what we previously discussed in Sect. 4.3, a match between LAT pulsar estimates and model prediction suggests that that model explains
the observed variety of γ-ray pulsar light curves.
The plane α vs. ζ as a function of the peak multiplicity and
peak separation and in the framework of PC, SG (two pole emission geometries), and OG (one pole emission geometry) was also
studied by Watters et al. (2009). Our study shows that the geometrical characteristics as a function of the pulsar orientation of
the light curves synthesised by PIERBA12 are consistent with
the simulations by Watters et al. (2009). Moreover our analysis
characterises the shoulder pulse shape that is recurrent among
both synthesised and observed objects. Shoulder shapes are observed in the framework of the SG and OG/OPC geometries; in
the SG model, the shoulder shape is associate with the α−ζ plane
region between single- and double-peak profiles, while in the
OG/OPC geometry they do not represent the transition between
single- and double-peak shapes but they are recurrent at high α
and high ζ. The best agreement between estimated parameters
and the bulk of the model prediction is observed in the framework of the OPC model but there are inconsistencies in the peak
separations plot (Fig. 14), with the OG and OPC model predicting too few small α values than the fits to the LAT pulsars. The
PC estimates completely fail to match the major part of the PC
simulations for all α and ζ. The SG estimates show inconsistencies in the light-curve peak multiplicity and pulsar orientation;
many two peak pulsars are found at high α and low ζ where four
peak profiles are found (see left side SG panel of Fig. 14). The
OG/OPC model estimates are consistent with model predictions
for what concern both multiplicity and peak separation. Other
plots showing α and ζ as a function of spin period (just for the
PC)/gap width (SG, OG, and OPC) and light-curve peak multiplicity, are shown in Figs. D.1 and D.2.
40
50
60
70
80
90
0
20
40
α [deg]
60
80
0
20
40
α [deg]
60
80
Fig. 14. From the first to the fourth row, the γ-ray light curve multiplicity (left column) and peak separation (right column) as a function of α
and ζ for LAT pulsars and PC, SG, OG, and OPC models are shown
respectively. Big triangles and small points refer to LAT and simulated
pulsars, respectively.
A137, page 15 of 26
A&A 588, A137 (2016)
Table 5. Estimates of the consistency between observed/estimated and simulated two-dimensional distributions shown in Figs. 7 to 13.
Fig. 7
Δγ vs. fΩ
Δradio vs. fΩ
Fig. 8
Δγ vs. α
Δradio vs. α
Fig. 9
Δγ vs. ζ
Δradio vs. ζ
Fig. 10
δ vs. fΩ
δ vs. α
δ vs. ζ
Fig. 11
Δγ vs. Ė
Δradio vs. Ė
δ vs. Ė
Fig. 12
Δγ vs. P
Δradio vs. P
δ vs. P
Fig. 13
Δγ vs. δ
Δradio vs. δ
D
PC
pvalue [%]
D
SG
pvalue [%]
D
OG
pvalue [%]
D
OPC
pvalue [%]
0.72
0.48
0.18
0.18
0.40
0.29
0.27
0.16
0.55
0.35
0.25
0.26
0.35
0.29
24
0.17
0.69
0.54
0.26
0.23
0.23
0.24
0.22
0.20
0.57
0.39
0.21
0.19
0.43
0.39
0.20
0.16
0.73
0.51
0.23
0.14
0.23
0.24
0.22
0.15
0.45
0.38
0.23
0.22
0.34
0.37
0.22
0.18
0.50
0.52
0.51
0.23
0.24
0.22
0.48
0.49
0.49
0.24
0.25
0.24
0.32
0.38
0.32
0.27
0.24
0.21
0.34
0.33
0.35
25
0.28
0.27
0.74
0.44
0.56
0.20
0.23
0.24
0.53
0.40
0.49
0.25
0.20
0.26
0.52
0.43
0.42
0.23
0.19
0.28
0.49
0.40
0.39
0.23
0.29
0.24
0.71
0.48
0.54
0.22
0.23
0.25
0.53
0.43
0.50
0.24
0.19
0.24
0.47
0.49
0.39
0.23
0.16
0.25
0.43
0.45
0.38
0.27
0.17
0.25
0.43
0.16
0.25
0.32
0.38
0.13
0.23
0.23
0.28
0.12
0.24
0.41
0.25
0.08
0.25
0.21
Notes. The 2D-2KS statistics D ranges between 0 and 1 for distributions showing total agreement and total disagreement, respectively. The pvalue
is the probability to obtain the observed D value under the assumption that the two distributions are obtained from the same distribution (null
hypothesis). This is equivalent to rejecting the null hypothesis at a confidence level of 100-(pvalue )%. The 2D-2KS statistics and distributions are
described in Sect. E. The D and pvalue parameters relative to the first and second most consistent distributions are highlighted in dark grey cells and
light grey cells, respectively.
6. Summary
We compared the morphological characteristics of observed
γ-ray and radio light curves with the same characteristics computed on a synthesised pulsar light-curve population in the
framework of different γ-ray and radio geometrical models.
The observed γ-ray and radio light curves are the young or
middle-aged ordinary pulsars published in Abdo et al. (2013),
while the simulated pulsar light curves are those synthesised
by Pierbattista et al. (2012) assuming the magnetospheric emission geometry from Dyks et al. (2004) and in the framework of
four γ-ray models and a radio model: Polar Cap (PC; Muslimov
& Harding 2003), Slot Gap (SG; Muslimov & Harding 2004),
Outer Gap (OG; Cheng et al. 2000), One Pole Caustic (OPC;
Watters et al. 2009; Romani & Watters 2010), and radio core plus
cone models (Gonthier et al. 2004; Story et al. 2007; Harding
et al. 2007).
For observed and simulated light curves and for each model
we defined a series of morphological characteristics, namely
peak number, light-curve minima, width of the peaks, among
others, and built a γ-ray and radio shape classification according
to the recurrence of these characteristics in the light curve. We
evaluated the precise peak phases by fitting the light curve with a
number of Gaussian and/or Lorentzian functions equal to lightcurve peak number and computed γ-ray and radio peak separation distributions (Δγ and ΔRadio , respectively) and the radio loud
pulsars, radio-radio lag distribution (δ) for observed objects and
for the objects simulated in the framework of each model.
We studied how Δγ , ΔRadio , and δ changes as a function of
observable pulsar characteristics, namely spin period (P) and
A137, page 16 of 26
spin-down power (Ė), and as a function of non-observable pulsar
characteristics like magnetic obliquity (α), observer line of sight
(ζ), and γ-ray beaming factor ( fΩ ). The observable pulsar parameters are taken from the second catalogue of γ-ray LAT pulsars,
(Abdo et al. 2013) while the non-observable LAT pulsar parameters were estimated in the framework of PC, SG, OG, and OPC
models by Pierbattista et al. (2015). We compared the observed
distributions of Δγ , ΔRadio , and δ with the same distributions obtained in the framework of the implemented γ-ray emission geometries. We also compared observed and simulated trends for
Δγ , ΔRadio , and δ as a function of the observable parameters P
and Ė, and as a function of the non-observable/estimated pulsar
parameters α, ζ, and fΩ . The comparison of the ΔRadio distribution and trends within each γ-ray models is possible because the
RL pulsar population changes as a function of the different γ-ray
visibility of each model and shows how the characteristics of a
unique radio population change in the framework of its γ rayselected subsamples.
We studied how the recurrence of the shape classes changes
between the whole simulated sample and its visible subsample
and we obtained that no selection effects due to the light-curve
shapes affect the pulsar visibility. This allowed us to compare
the observed morphological characteristics with the same characteristics obtained on the whole simulated sample and within
each model without applying the LAT γ-ray pulsar visibility to
the simulated pulsar sample.
The agreement between the observed and simulated onedimensional distribution in the framework of each model has
been quantified by computing the 2KS test. Each 2KS statistics D, has been computed jointly with its pvalue expressing the
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
probability of obtaining the D value assuming that the two onedimensional distribution are obtained from the same underlying distribution (null hypothesis). The agreement between the
observed/estimated and simulated two-dimensional distributions
has been quantified by computing the two-dimensional 2KS
statistics D with relative pvalue through bootstrap resampling, as
described in Sect. E.
We obtain that none of the proposed emission geometries
explains the LAT pulsar morphology. The OPC model provides,
overall, a best explanation of the observed morphological characteristics and trends but the two-pole caustic emission from the
SG is necessary to explain some important morphological characteristics. Comparisons of observed and simulated γ-ray peak
multiplicity (number of light curve peaks) show that the OPC
model manages to explain the observed light-curve complexity.
We studied how the γ-ray peak multiplicity changes from RQ
to RL pulsars and we observed an increase of the single-peak,
γ-ray light curves among RL objects that is not explained in the
framework of the implemented emission geometries.
The comparison of observed and simulated γ-ray and radio
peak separations show that the SG and OPC models explain the
observations in a complementary way. The SG best explain the
observed wide separated peaks but do not manage to explain
the observation at low peak separation while the OPC gives a
good explanation of the low peak-separation region but completely fails to explain wide peak separation. This suggest that
the OPC model best explains the structure of the γ-ray peak but
the two-pole caustic SG model is required to explain the 0.5 separated features generated by the emission from the two magnetic poles appearing in the light curves at high α and ζ. Among
the γ ray-selected radio pulsars, all the models explain the radio peak-separation distribution but the SG model gives the best
description of the observed trend. Overall, none of the assumed
emission geometries explain the observed radio-lag distribution
with all models underestimating wide radio lag, δ > 0.4. The
OPC model best explains the observed distribution since it predicts the lack of measures for δ ∼ 0.09 and partly explains the
peak observed at δ ∼ 0.15.
The comparison of simulated and observed trends for Δγ ,
ΔRadio , and δ as a function of the non-observable pulsar parameters suggests a higher agreement between observation and estimations in the framework of the OPC. In all studied cases, the
OPC and SG estimated trends show a best match with the modelled trends with the OPC estimates best matching the model
predictions for Δγ vs. fΩ , ΔRadio vs. α, Δγ vs. ζ, and ΔRadio vs. α,
and particularly for δ vs. α, ζ, and fΩ . This confirms that the OPC
model predicts the ranges of light-curve shapes that best explains
the LAT findings. The comparison of simulated and observed
trends for Δγ , ΔRadio , and δ as a function of P and Ė confirms the
lack of high Ė objects found, for all models, by Pierbattista et al.
(2012). The observed lack affects all simulated distributions that
in no case explains the LAT findings.
In the plane δ − Δγ and δ − ΔRadio , the only model that correctly explains the observations is the OPC. The δ − Δγ plane is a
tracker of the magnetospheric structure of the γ-ray gap emission
region and of the relative position of γ-ray and radio emission regions. This allows us to conclude that, under the assumption that
our radio core plus cone emission geometry is correct for the
major part of the LAT pulsars, the outer magnetosphere of a pulsar is the most likely location for γ-ray photon production. The
δ − ΔRadio plane is a tracker of the radio emission beam structure
and of its position with respect to the γ-ray emission region and
the OPC model gives the best explanation of the observed trend.
We obtained a map of peak multiplicity and Δγ as a function of α and ζ and studied for which model, the α, ζ estimated
for LAT pulsars match the model prediction for the same peak
multiplicity and peak separation. We obtain that the OPC model
prediction for LAT pulsars α and ζ best match the simulations
for the same peak multiplicity and Δγ .
Despite the larger agreement between the OPC model prediction and observation, one has to note that δ is strongly dependent on the assumed magnetospheric structure that in this study
is a VRD. Kalapotharakos et al. (2012) has shown that δ is larger
in DMs and increases with increasing conductivity from 0 (VRD
magnetosphere) to ∞ (FFE magnetosphere). We can expect that
the same model predictions obtained in the framework of a nonVRD magnetosphere (conductivity larger than 0) would imply
larger values of δ that would modify the δ distribution and the
Δγ as a function of δ in the framework of the implemented γ-ray
models. A δ increase has already been shown in the OG model
prediction by Kalapotharakos et al. (2012) and would surely increase the agreement between the SG prediction and observations. Particularly interesting is the model recently proposed by
Kalapotharakos et al. (2014), which assumes a FFE magnetosphere with an infinite conductivity, inside the light cylinder and
a DM with finite but high conductivity, outside the light cylinder
(FFE Inside, Dissipative Outside, FIDO). The FIDO model best
explains the observed δ and Δγ distributions and predict the correct variation of Δγ as a function of δ merging characteristics of
both SG and OG models.
The results obtained by Kalapotharakos et al. (2012) and
by Kalapotharakos et al. (2014), in conjunction with the results of the current study, suggest a complementary nature of
SG and OPC in explaining different aspects of the observed phenomenology, support the existence of a hybrid SG/OG emission
mechanism and a non-VRD magnetosphere geometry, and calls
for a more physical explanation of the OPC model to better understand the pulsar magnetosphere.
Acknowledgements. The Fermi LAT Collaboration acknowledges generous ongoing support from a number of agencies and institutes that have supported
both the development and operation of the LAT as well as scientific data analysis. These include the National Aeronautics and Space Administration and
the Department of Energy in the United States, the Commissariat à l’Énergie
Atomique and the Centre National de la Recherche Scientifique/Institut National
de Physique Nucléaire et de Physique des Particules in France, the Agenzia
Spaziale Italiana and the Istituto Nazionale di Fisica Nucleare in Italy, the
Ministry of Education, Culture, Sports, Science and Technology (MEXT),
High Energy Accelerator Research Organization (KEK) and Japan Aerospace
Exploration Agency (JAXA) in Japan, and the K. A. Wallenberg Foundation, the
Swedish Research Council and the Swedish National Space Board in Sweden.
Additional support for science analysis during the operations phase is gratefully acknowledged from the Istituto Nazionale di Astrofisica in Italy and the
Centre National d’Études Spatiales in France. M.P. acknowledges the Nicolaus
Copernicus Astronomical Center, grant DEC-2011/02/A/ST9/00256, for providing software and computer facilities needed for the development of this work.
M.P. gratefully acknowledges Eric Feigelson for useful discussions and suggestions. P.L.G. thanks the National Science Foundation through Grant No.
AST-1009731 and the NASA Astrophysics Theory Program through Grant No.
NNX09AQ71G for their generous support. The authors gratefully acknowledge the Pulsar Search and Timing Consortia, all the radio scientists who contributed in providing the radio light curves used in this paper, and the radio
observatories that generated the radio profiles used in this paper: the Parkes
Radio Telescope is part of the Australia Telescope which is funded by the
Commonwealth Government for operation as a National Facility managed by
CSIRO; the Green Bank Telescope is operated by the National Radio Astronomy
Observatory, a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc; the Arecibo Observatory
is part of the National Astronomy and Ionosphere Center (NAIC), a national
research center operated by Cornell University under a cooperative agreement
with the National Science Foundation; the Nançay Radio Observatory is operated by the Paris Observatory, associated with the French Centre National de la
A137, page 17 of 26
A&A 588, A137 (2016)
Recherche Scientifique (CNRS); the Lovell Telescope is owned and operated by
the University of Manchester as part of the Jodrell Bank Centre for Astrophysics
with support from the Science and Technology Facilities Council of the United
Kingdom; the Westerbork Synthesis Radio Telescope is operated by Netherlands
Foundation for Radio Astronomy, ASTRON.
References
Abdo, A. A., Ackermann, M., Ajello, M., et al. 2010, ApJ, 187, 460
Abdo, A. A., Ajello, M., Allafort, A., et al. 2013, ApJS, 208, 17
Babu, G., & Rao, C. 1993, Bootstrap methodology. Handbook of Statistics 9
(Elsevier B.V.), 627
Bai, X., & Spitkovsky, A. 2010, ApJ, 715, 1282
Cheng, K. S., Ruderman, M., & Zhang, L. 2000, ApJ, 537, 964
Dyks, J., Harding, A. K., & Rudak, B. 2004, ApJ, 606, 1125
Gonthier, P. L., Van Guilder, R., Harding, A. K., Greneir, I., & Perrot, C. 2004,
BAAS, 36, 918
Harding, A. K., Grenier, I. A., & Gonthier, P. L. 2007, Ap&SS, 309, 221
Kalapotharakos, C., Harding, A. K., Kazanas, D., & Contopoulos, I. 2012, ApJ,
754, L1
Kalapotharakos, C., Harding, A. K., & Kazanas, D. 2014, ApJ, 793, 97
A137, page 18 of 26
Muslimov, A. G., & Harding, A. K. 2003, ApJ, 588, 430
Muslimov, A. G., & Harding, A. K. 2004, ApJ, 606, 1143
Pierbattista, M. 2010, Ph.D. Thesis, Université Paris 7 Denis Diderot
[arXiv:1309.5982]
Pierbattista, M., Grenier, I. A., Harding, A. K., & Gonthier, P. L. 2012, A&A,
545, A42
Pierbattista, M., Harding, A. K., Grenier, I. A., et al. 2015, A&A, 575, A3
Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery, B. P. 1992,
Numerical Recipes: The Art of Scientific Computing, 2nd edn. (Cambridge:
Cambridge University Press)
Ray, P. S., Abdo, A. A., Parent, D., et al. 2012, 2011 Fermi Symp. Proc., eConf
C110509 [arXiv:1205.3089]
Romani, R. W., & Watters, K. P. 2010, ApJ, 714, 810
Romani, R. W., & Yadigaroglu, I.-A. 1995, ApJ, 438, 314
Smith, D. A., Guillemot, L., Camilo, F., et al. 2008, A&A, 492, 923
Starck, J.-L., Pires, S., & Réfrégier, A. 2006, A&A, 451, 1139
Story, S. A., Gonthier, P. L., & Harding, A. K. 2007, ApJ, 671, 713
Takata, J., Wang, Y., & Cheng, K. S. 2011, ApJ, 726, 44
Venter, C., Johnson, T. J., & Harding, A. K. 2012, ApJ, 744, 34
Watters, K. P., & Romani, R. W. 2011, ApJ, 727, 123
Watters, K. P., Romani, R. W., Weltevrede, P., & Johnston, S. 2009, ApJ, 695,
1289
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
Appendix A: The γ-ray and radio light-curve shape classification
Table A.1. Shape classification and peak multiplicity for simulated and observed γ-ray light curves as defined in Table 2. The empty cell in
correspondence of a shape class for a particular model indicates that this light-curve shape is not observed in the framework of this model.
Shape Class
Multiplicity
1- Bump
1
2- Sharp
1
3- Shoulder
1
4- Two
2
5- Double
2
6- Double+Single
3
7- Triple/Three
3
8- Two double
4
PC Light curve
SG Light curve
OG Light curve
OPC Light curve
A137, page 19 of 26
A&A 588, A137 (2016)
Table A.2. Shape classification and peak multiplicity for simulated and observed radio light curves as defined in Table 3.
Shape Class
A137, page 20 of 26
Multiplicity
1- Bump
1
2- Sharp
1
3- Two
2
4- Double
2
5- Double+Single
3
6- Two Double
4
7- Triple
3
8- Three
3
9- Two Triple
4
Radio Core plus cone
and Radio LAT
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
Appendix B: The pulsar γ-ray and radio emission pattern
PC
OG/OPC
SG
50
10
0
50
20
0
50
30
0
50
40
ζ [◦]
0
50
50
0
50
60
0
50
70
0
50
80
0
50
0
-0.5
90
0. 0
0.5
-0.5
0. 0
-0.5
0.5
0.0
0.5
Pulse phase
Fig. B.1. For each γ-ray emission model, the pulsar γ-ray (shaded surface) and radio (blue contours) emission patterns (phase plots) as a function
of magnetic obliquity α, are shown. Each phase-plot panel gives the pulsar light curve as a function of the observer line-of-sight ζ for each α value
stepped every 10◦ in the interval 0◦ < α < 90◦ . The γ-ray and radio phase plots have been obtained for a magnetic field strength of 108 Tesla and
spin period of 30 ms for the PC and radio cases, and gap widths of 0.04 and 0.01 for the SG and OG/OPC cases, respectively.
A137, page 21 of 26
A&A 588, A137 (2016)
Appendix C: Recurrence of γ-ray and radio shape classes
Figure C.1 compares simulated and observed multiplicities of γ-ray-selected radio profiles in the framework of each model. The radio model is
unique, but the RL pulsars subsample changes with each model γ-ray visibility generating different radio-peak multiplicity distributions. Figure C.2
compares simulated and observed γ-ray peak multiplicity for RQ and RL pulsars in the framework of each model. Figure C.3 compares the
recurrence of the shape classes defined in Tables 2 and 3 and shown in Tables A.1 and A.2 for the simulated γ-ray and radio pulsars, respectively.
The statistical agreement between observed and simulated one-dimensional distributions shown in Figs. C.1 to C.3 are listed in Table C.1.
Table C.1. Two-sample Kolmogorov-Smirnov statistics (2KS) and relative pvalue between observed and simulated one-dimensional distributions
shown in Figs. C.1 to C.3 and each model.
Fig. C.1: Radio peak multiplicity distribution
Fig. C.2-left: RQ γ-ray peak multiplicity distribution
Fig. C.2-right: RL γ-ray peak multiplicity distribution
Fig. C.3-left: RQ shape-classes distribution
Fig. C.3-right: RL shape-classes distribution
D
PC
pvalue [%]
D
SG
pvalue [%]
D
OG
pvalue [%]
D
OPC
pvalue [%]
0.42
0.72
0.36
0.72
0.36
3e-8
2.4
1.8
2.4
1.8
0.37
0.38
0.35
0.38
0.40
5e-6
2e-2
6e-3
2e-2
3e-4
0.45
0.17
0.41
0.17
0.59
1e-9
33
1e-4
33
2e-11
0.417
0.05
0.22
0.18
0.56
4e-8
99.9
2.6
24
2e-10
Notes. The 2KS statistics ranges between 0 and 1 for distributions showing total agreement and total disagreement, respectively. The pvalue is the
probability to obtain observed 2KS value under the assumption that the two distributions are obtained from the same distribution (null hypothesis).
This is equivalent to reject the null hypothesis at a confidence level of 100-(pvalue ). The 2KS test is described in Sect. E. The D parameters relative
to the first and second most consistent distributions are highlighted in dark grey and light grey cells, respectively.
% of the total
PC γ -ray and radio pulsars
SG γ -ray and radio pulsars
80
80
60
60
40
40
20
20
0
0
% of the total
OG γ -ray and radio pulsars
OPC γ -ray and radio pulsars
80
80
60
60
40
40
20
20
0
0
1
2
3
4
5
6
1
2
3
4
5
6
Radio peak multiplicity
Radio peak multiplicity
Fig. C.1. Recurrence of peak multiplicities for the simulated radio pulsar populations (grey) and LAT population (red) and each model. The radio
peak multiplicities are defined in Table A.1.
PC RQ γ -ray pulsars
80
SG RQ γ -ray pulsars
80
PC RL γ -ray pulsars
SG RL γ -ray pulsars
60
% of the total
% of the total
60
60
60
40
40
20
20
0
0
30
30
20
20
10
10
0
OPC RQ γ -ray pulsars
OG RL γ -ray pulsars
% of the total
60
60
40
40
20
OPC RL γ -ray pulsars
60
80
% of the total
40
40
0
OG RQ γ -ray pulsars
80
50
50
20
80
50
40
60
30
40
20
20
10
0
0
1
1s
2
3
γ -ray peak multiplicity
4
0
1
1s
2
3
γ -ray peak multiplicity
4
0
1
1s
2
3
γ -ray peak multiplicity
4
1
1s
2
3
γ -ray peak multiplicity
4
Fig. C.2. Recurrence of the γ-ray peak multiplicity for the RQ and RL pulsars of simulated (grey) and observed (red) populations and each model
are shown in the top and bottom panels, respectively.
A137, page 22 of 26
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
SG γ -ray pulsars
40
60
30
20
40
10
20
0
0
OG γ -ray pulsars
% of the total
60
50
40
20
30
20
10
0
1
2
3
4
5
6
γ -ray classes
7
8
50
50
40
40
30
30
20
20
10
10
0
OPC γ -ray and radio pulsars
60
60
50
50
40
40
30
30
20
20
10
10
0
60
OG γ -ray and radio pulsars
70
30
SG γ -ray and radio pulsars
60
0
OPC γ -ray pulsars
40
% of the total
PC γ -ray and radio pulsars
% of the total
% of the total
PC γ -ray pulsars
80
1
2
3
4
5
6
γ -ray classes
7
8
10
0
0
1
2
3
4
5
6
Radio classes
7
8
1
2
3
4
5
6
7
8
Radio classes
Fig. C.3. Recurrence shape classes for simulated γ-ray and radio pulsar populations (grey) and LAT population (red) and each model are shown in
the top and bottom panel, respectively.
A137, page 23 of 26
A&A 588, A137 (2016)
Appendix D: Multiplicities and peak separations as a function of P and w in the α − ζ plane
PC
0
l o g1 0 ( P [ s )] ≤ − 0 . 8
SG
OG
OPC
w S G≤ 0 . 1
w O G≤ 0 . 1
w O P C≤ 0 . 1
10
ζ [ d e g]
20
30
40
50
60
70
80
90
0
− 0 . 8 < l o g1 0 ( P [ s )] ≤ − 0 . 6
0 . 1 < w S G≤ 0 . 2
0 . 1 < w O G≤ 0 . 2
0 . 1 < w O P C≤ 0 . 2
− 0 . 6 < l o g1 0 ( P [ s )] ≤ − 0 . 4
0 . 2 < w S G≤ 0 . 3
0 . 2 < w O G≤ 0 . 3
0 . 2 < w O P C≤ 0 . 3
− 0 . 4 < l o g1 0 ( P [ s )] ≤ − 0 . 2
0 . 3 < w S G≤ 0 . 4
0 . 3 < w O G≤ 0 . 4
0 . 3 < w O P C≤ 0 . 4
10
ζ [ d e g]
20
30
40
50
60
70
80
90
0
10
ζ [ d e g]
20
30
40
50
60
70
80
90
0
10
ζ [ d e g]
20
30
40
50
60
70
80
90
0
l o g1 0 ( P [ s )] > − 0 . 2
w S G> 0 . 4
w O G> 0 . 4
w O P C> 0 . 4
10
ONE
S H OU L DER
T WO
T H R EE
FOUR
ζ [ d e g]
20
30
40
50
60
70
80
90
0
20
40
60
α [ d e g]
80
0
20
40
60
α [ d e g]
80
0
20
40
60
α [ d e g]
80
0
20
40
60
α [ d e g]
80
Fig. D.1. Peak multiplicity as a function of spin period (PC) and gap width (SG, OG, OPC) in the α − ζ plane. Left to right columns refer to PC, SG,
OG, and OPC, respectively while top to bottom rows refer to increasing period (PC) and gap width (SG, OG, OPC). This figure can be compared
with similar figures of Watters et al. (2009).
A137, page 24 of 26
M. Pierbattista et al.: Young and middle age pulsar light-curve morphology
PC
0
l o g1 0 ( P [ s )] ≤ − 0 . 8
SG
OG
OPC
w S G≤ 0 . 1
w O G≤ 0 . 1
w O P C≤ 0 . 1
10
ζ [ d e g]
20
30
40
50
60
70
80
90
0
− 0 . 8 < l o g1 0 ( P [ s )] ≤ − 0 . 6
0 . 1 < w S G≤ 0 . 2
0 . 1 < w O G≤ 0 . 2
0 . 1 < w O P C≤ 0 . 2
− 0 . 6 < l o g1 0 ( P [ s )] ≤ − 0 . 4
0 . 2 < w S G≤ 0 . 3
0 . 2 < w O G≤ 0 . 3
0 . 2 < w O P C≤ 0 . 3
− 0 . 4 < l o g1 0 ( P [ s )] ≤ − 0 . 2
0 . 3 < w S G≤ 0 . 4
0 . 3 < w O G≤ 0 . 4
0 . 3 < w O P C≤ 0 . 4
10
ζ [ d e g]
20
30
40
50
60
70
80
90
0
10
ζ [ d e g]
20
30
40
50
60
70
80
90
0
10
ζ [ d e g]
20
30
40
50
60
70
80
90
0
l o g1 0 ( P [ s )] > − 0 . 2
w S G> 0 . 4
w O G> 0 . 4
w O P C> 0 . 4
10
0. 0- 0.1
0. 1- 0.2
0. 2- 0.3
0. 3- 0.4
> 0. 4
ζ [ d e g]
20
30
40
50
60
70
80
90
0
20
40
60
α [ d e g]
80
0
20
40
60
α [ d e g]
80
0
20
40
60
α [ d e g]
80
0
20
40
60
α [ d e g]
80
Fig. D.2. Peak separation as a function of spin period (PC) and gap width (SG, OG, OPC) in the α − ζ plane. Left to right columns refer to PC, SG,
OG, and OPC, respectively while top to bottom rows refer to increasing period (PC) and gap width (SG, OG, OPC). This figure can be compared
with similar figures of Watters et al. (2009).
A137, page 25 of 26
A&A 588, A137 (2016)
Appendix E: Two-sample Kolmogorov-Smirnov test
The two-sample Kolmogorov-Smirnov test (2KS) is a nonparametric test that allows statistical quantification of the degree
of agreement of two one-dimensional distributions. By computing the maximum distance between the cumulative density functions (CDFs) of the tested distributions, the 2KS test allows for
the rejection of the null hypothesis: that the distributions are obtained from the same underlying distribution at a predefined confidence level (CL). We consider two samples, A and B, of sizes a
and b, respectively. If A(x) and B(x) are the CDF of the samples
A and B, respectively, the 2KS statistics is given by
Da,b = sup |A(x) − B(x)|
(E.1)
x
where sup is the supremium function. The null hypothesis that
the samples A and B have been obtained from the same underlying distribution is rejected at CL pvalue if
a+b
(E.2)
Da,b > c(pvalue)
ab
with c(pvalue ) a tabulated value function of the chosen CL. The D
value is included between 0 and 1 indicating total agreement and
total disagreement between the tested distributions, respectively.
The pvalue gives the probability to observe the relative statistics
Da,b assuming the null hypothesis to be true. The 2KS statistic
and pvalue given in Table 4 and relative to one-dimension observed and simulated distributions shown in Figs. 4 to 6 have
been computed with Eqs. (E.1) and (E.2), respectively.
E.1. Using the 2KS to quantify the agreement between two
two-dimensional distributions
We used the two-dimensional version of the 2KS statistics (2D2KS) described in Press et al. (1992) to quantify the agreement
A137, page 26 of 26
between observed/estimated and simulated two-dimensional distributions shown in Figs. 7 to 13. The 2D-2KS statistics described in Press et al. (1992) gives a reasonable estimate of the
distance between simulated and observed 2D distributions but,
given the high number of possible ways to sort the 2D data
plane, the 2D-2KS statistics is not distribution free as its 1D version, and cannot be associated with the probability that observations and simulations are obtained from the same parent distribution (pvalue ). In order to evaluate the pvalue associated with
the 2D-2KS statistics, we studied the statistical distribution of
the 2D-2KS statistics by paired bootstrap resampling of the observed 2D sample, and use it to evaluate the pvalue corresponding
to each 2D-2KS statistics.
The paired bootstrap resampling consists in resampling the
observed 2D distribution by randomly rearrange the observed
values on the 2D plane. It is possible to build a high number
of bootstrapped distribution, statistically consistent with the observed 2D distribution, and use them to study the statistical distribution of a parameter of the observed 2D distribution (e.g. the
variance). An accurate description of the paired bootstrap resampling can be found in Babu & Rao (1993).
We used the paired bootstrap method to resample each observed 2D distribution and create a large number of bootstrapped
samples. We computed the 2D-2KS statistics between each bootstrapped sample and the simulated sample in object, and we
built the statistical distribution of their 2D-2KS statistics for each
model. Once we evaluated the statistical distribution of the 2D2KS statistics, we used it to read off the probability pvalue corresponding to the 2D-2KS statistics computed between the simulated and observed/estimated 2D distribution in the object.